AllyshipContent Radar

News 2897

Chủ đề cộng đồng đang thảo luận

Crawl live-1795126500000 · 05:15 20 thg 11, 2026 · Xuất bản 05:31 20 thg 11, 2026

103 / 2897 chủ đề đạt ngưỡng 55 trong khoảng này.

1

@badlogicgames his approach is amazing. it's in fact so amazing that i made an entire framework around that approach a year ago

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

devagrawal09 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

@badlogicgames his approach is amazing. it's in fact so amazing that i made an entire framework around that approach a year ago https://devagrawal09.github.io/specter/

2

Amazon SageMaker AI now detects latency drift, idle GPUs, and missed optimizations on your inference endpoints

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Hazim Qudah 1 bài viết · AWS Machine Learning Blog
Nguyên văn nguồn
Chi tiết nguồn

<p>Every inference endpoint is well tuned the day it goes into production. Teams choose an instance type, serving configuration, and scaling settings that suit the model and its traffic at launch. With the endpoint serving smoothly, they understandably move on to the next priority. The workload, however, rarely stays still. Traffic patterns evolve, prompts grow longer, serving engines such as vLLM ship new releases, SageMaker inference develops new optimization techniques, and new fine-tuned checkpoints change the resources your model needs. Over time, a well-tuned endpoint quietly drifts. Latency and time to first token edge upward. A configuration that was never quite ideal keeps missing available performance gains. A GPU provisioned for demand that never materialized sits idle week after week. Because nothing has failed, no alarm sounds, and the drift goes unnoticed until users feel it or the bill reflects it.</p> <p><a href="https://aws.amazon.com/sagemaker/ai/" target="_blank" rel="noopener">Amazon SageMaker AI</a> now monitors your generative AI inference endpoints and tells you when a change could improve performance or lower cost by using <a href="https://aws.amazon.com/blogs/machine-learning/amazon-sagemaker-ai-now-supports-optimized-generative-ai-inference-recommendations/" target="_blank" rel="noopener">Generative AI optimized inference recommendations</a>&nbsp;feature with Agentic experience in &nbsp;<a href="https://aws.amazon.com/sagemaker/ai/studio/" target="_blank" rel="noopener">Amazon SageMaker Studio</a>. Monitoring is on by default for SageMaker AI inference endpoints using vLLM serving containers and GPU instances, with no setup required, and you can opt out of endpoints. SageMaker AI looks only at operational signals: latency, GPU utilization, and the endpoint’s serving configuration. It does not read your requests or responses. From those signals, it surfaces three kinds of findings:</p> <ul> <li><strong>Latency drift:</strong> A week-over-week rise in median (p50) end-to-end latency or time to first token.</li> <li><strong>Unused capacity:</strong> GPUs you pay for that sit mostly idle.</li> <li><strong>Configuration gaps:</strong> A serving container that’s missing newer performance improvements or security patches, or an optimization such as speculative decoding or GPU kernel tuning that your model supports but the endpoint doesn’t use.</li> </ul> <p>SageMaker AI sends a finding through AWS Health, with one notification per endpoint that covers that endpoint’s findings. When a notification arrives, you can act on it in one place. It opens the SageMaker AI agentic experience in Amazon SageMaker Studio with the endpoint, and its recent traffic already filled in. From there, you run inference recommendations to benchmark your current configuration against alternatives and compare them side by side on response time, throughput, and cost per million tokens. That turns a judgment call into a decision backed by data. Benchmarks run on separate compute, and nothing changes on your endpoint until you choose to apply a result.</p> <p>In this post, we show how SageMaker AI monitors your endpoints and how to read the notifications it sends through AWS Health. We then show how to go from a notification to a benchmarked decision in SageMaker Studio. We walk through three example scenarios, show how to apply or dismiss a recommendation safely, and explain how to opt an endpoint out of monitoring.</p> <h2 id="solution-overview">Solution overview</h2> <p>SageMaker AI inference recommendations follow a four-step loop: observe, notify, recommend, and decide. Each step is designed so that you stay in control, and your production endpoint isn’t changed without your explicit action.</p> <ol type="1"> <li><strong>Observe –</strong> Monitoring of latency and capacity usage is on by default for eligible endpoints. It doesn’t read your requests or responses, and it carries no additional GPU cost.</li> <li><strong>Notify –</strong>&nbsp;When there is something worth acting on, you receive a plain language notification in AWS Health, plus email, explaining what was observed, why it matters, and a recommended next step.</li> <li><strong>Recommend (opt in) –</strong>&nbsp;One selection opens the SageMaker AI agentic experience in SageMaker Studio, prefilled with the endpoint and its observed traffic, so you can benchmark alternatives on separate compute.</li> <li><strong>Decide –</strong>&nbsp;Compare response time, throughput, and cost side by side. Production is not changed until you apply a result.</li> </ol> <p>The feature produces a few types of findings. The following table summarizes what triggers each one and what gets tested when you choose to benchmark.</p> <table border="1px" width="100%" cellpadding="10px"> <tbody> <tr> <td><strong>Finding</strong></td> <td><strong>What triggers it</strong></td> <td><strong>What a benchmark tests</strong></td> </tr> <tr> <td>Latency drift</td> <td>Response time moves relative to the endpoint’s week-over-week trends</td> <td>Tensor parallelism settings and alternate instance types or sizes</td> </tr> <tr> <td>Underutilized GPUs</td> <td>GPU utilization stays well below capacity over the window</td> <td>Cheaper instance types and smaller sizes at observed peak traffic</td> </tr> <tr> <td>Speculative decoding not enabled</td> <td>Model and container support it, but it is off (configuration based)</td> <td>Current configuration compared to speculative decoding enabled</td> </tr> <tr> <td>Serving container</td> <td>The endpoint’s serving container has a newer version with a performance improvement and security patches for critical and high finding Common Vulnerabilities and Exposures (CVEs).</td> <td>Current compared to updated container</td> </tr> <tr> <td>Untuned GPU kernels</td> <td>Kernel configuration is not tuned for the model and hardware</td> <td>Default compared to tuned kernels</td> </tr> </tbody> </table> <h2 id="prerequisites">Prerequisite

3

Inside Amazon Connect Talent: The science of structured AI interviews

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Sunitha Srivatsa 1 bài viết · AWS Machine Learning Blog
Nguyên văn nguồn
Chi tiết nguồn

<p><span class="TextRun SCXW252313659 BCX0" lang="EN-US" xml:lang="EN-US" data-contrast="auto"><span class="NormalTextRun SCXW252313659 BCX0">For organizations hiring </span><span class="NormalTextRun SCXW252313659 BCX0">hundreds or </span><span class="NormalTextRun SCXW252313659 BCX0">thousands of people, every interview presents the same challenge: How do you give candidates a consistent opportunity to </span><span class="NormalTextRun SCXW252313659 BCX0">demonstrate</span><span class="NormalTextRun SCXW252313659 BCX0"> their skills while gathering the evidence needed to make informed hiring decisions?</span></span><span class="EOP Selected SCXW252313659 BCX0" data-ccp-props="{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:278}">&nbsp;</span></p> <p><span class="TextRun SCXW193008649 BCX0" lang="EN-US" xml:lang="EN-US" data-contrast="auto"><span class="NormalTextRun SCXW193008649 BCX0">Structured behavioral interviews offer a research-backed approach to this challenge. They use consistent, job-related questions and clear scoring criteria to give candidates an opportunity to </span><span class="NormalTextRun SCXW193008649 BCX0">demonstrate</span><span class="NormalTextRun SCXW193008649 BCX0"> their skills. Decades of research in industrial-organizational (I-O) psychology have shown that structured interviews help organizations evaluate candidates and predict job performance (Schmidt &amp; Hunter, 1998; Sackett et al., 2022). Yet delivering </span><span class="NormalTextRun SCXW193008649 BCX0">structured interviews</span><span class="NormalTextRun SCXW193008649 BCX0"> at scale requires </span><span class="NormalTextRun SCXW193008649 BCX0">significant time</span><span class="NormalTextRun SCXW193008649 BCX0"> and </span><span class="NormalTextRun SCXW193008649 BCX0">expertise</span><span class="NormalTextRun SCXW193008649 BCX0"> to develop</span><span class="NormalTextRun SCXW193008649 BCX0"> competencies, </span><span class="NormalTextRun SCXW193008649 BCX0">create q</span><span class="NormalTextRun SCXW193008649 BCX0">uestions, train interviewers, and evaluate responses consistently</span><span class="NormalTextRun SCXW193008649 BCX0">.</span></span><span class="EOP Selected SCXW193008649 BCX0" data-ccp-props="{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:278}">&nbsp;</span></p> <p><span data-contrast="auto"><a href="https://aws.amazon.com/products/connect/talent/" target="_blank" rel="noopener">Amazon Connect Talent</a> <span class="NormalTextRun SCXW175523192 BCX0">brings established hiring science to AI-led interviews at scale. </span><span class="NormalTextRun SCXW175523192 BCX0">Connect Talent </span><span class="NormalTextRun SCXW175523192 BCX0">uses structured questions, adaptive follow-ups, and evidence-based scoring to help organizations gather consistent, job-related information while giving more candidates the opportunity to demonstrate their strengths.</span></span></p> <p><span class="TextRun SCXW66502508 BCX0" lang="EN-US" xml:lang="EN-US" data-contrast="auto"><span class="NormalTextRun SCXW66502508 BCX0">We built Connect Talent around three core design principles and evaluated how well the service </span><span class="NormalTextRun ContextualSpellingAndGrammarErrorV2Themed SCXW66502508 BCX0">applies</span><span class="NormalTextRun SCXW66502508 BCX0"> them through </span><span class="NormalTextRun SCXW66502508 BCX0">a combination of </span><span class="NormalTextRun SCXW66502508 BCX0">internal research and independent expert review. </span><span class="NormalTextRun SCXW66502508 BCX0">The </span><span class="NormalTextRun SCXW66502508 BCX0">findings offer a closer look at the science behind </span><span class="NormalTextRun SCXW66502508 BCX0">Connect Talent’s </span><span class="NormalTextRun SCXW66502508 BCX0">AI-le</span><span class="NormalTextRun SCXW66502508 BCX0">d,</span><span class="NormalTextRun SCXW66502508 BCX0"> structured interviewing.</span></span><span class="EOP Selected SCXW66502508 BCX0" data-ccp-props="{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:278}">&nbsp;</span></p> <h2><b><span data-contrast="auto">Three core design principles</span></b></h2> <h3><b><span data-contrast="auto">1. Build interviews around job-related evidence</span></b></h3> <p><span class="TextRun SCXW227730746 BCX0" lang="EN-US" xml:lang="EN-US" data-contrast="auto"><span class="NormalTextRun SCXW227730746 BCX0">Connect Talent builds each interview around the competencies </span><span class="NormalTextRun SCXW227730746 BCX0">required</span><span class="NormalTextRun SCXW227730746 BCX0"> for the role. The service uses job requirements and </span><span class="NormalTextRun ContextualSpellingAndGrammarErrorV2Themed SCXW227730746 BCX0">established</span><span class="NormalTextRun SCXW227730746 BCX0"> occupational frameworks to create behaviorally anchored scorecards. Each scorecard defines specific strength and concern signals, and every interview question targets the behaviors it aims to uncover.</span></span><span class="EOP Selected SCXW227730746 BCX0" data-ccp-props="{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:278}">&nbsp;</span></p> <p><span class="TextRun SCXW128739577 BCX0" lang="EN-US" xml:lang="EN-US" data-contrast="auto"><span class="NormalTextRun SCXW128739577 BCX0">The AI interviewer also adapts to each response. After every answer, it checks whether the candidate has provided enough evidence for the competency being assessed. When details are missing, it asks targeted follow-up questions. This approach preserves the structure of the interview while giving candidates more opportunities to explain their experience.</span></span><span class="EOP Selected SCXW128739577 BCX0" data-ccp-props="{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:278}">&nbsp;</span></p> <h3><b><span data-contrast="auto">2. Separate interviewing from scoring</span></b></h3> <p><span c

4

Genie 4 is the first model we launched on the new Live Interactions API, bringing native WebRTC (with ephemeral tokens) and WebSocket support directly to Interactions and Gemini API.

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

_philschmid 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Genie 4 is the first model we launched on the new Live Interactions API, bringing native WebRTC (with ephemeral tokens) and WebSocket support directly to Interactions and Gemini API. This is part of our ongoing work to improve Gemini API experience. When we started building the Interactions API, one of our main goals was making sure standard API calls and real-time and long running sessions share one consistent interface. We put a lot of effort into getting the primitives right, and we have big plans. I would love your feedback as we build this out: what do you want to see as we bring protocols like WebSockets and WebRTC to more Gemini models? What is still missing for building better agent experiences or real-time apps? In the meantime, you can try Genie 4 in the docs: https://aistudio.google.com/docs/genie Media

5

Watch → https://goo.gle/4AIPYYS

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Google DeepMind 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Watch → https://goo.gle/4AIPYYS Spotify → https://goo.gle/4iSB5x5 Apple Podcasts → https://goo.gle/4iSrk21 Or listen wherever you get your podcasts! 🎧

6

Our SVP @koraykv believes there won't be a single test or benchmark that tells us we've reached AGI. Instead, progress looks like models becoming less spiky and more reliable across our daily lives.

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Google DeepMind 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Our SVP @koraykv believes there won't be a single test or benchmark that tells us we've reached AGI. Instead, progress looks like models becoming less spiky and more reliable across our daily lives. Listen in as he and @fryrsquared discuss the road ahead ↓ Timecodes: 00:00 Introduction 01:26 Evolution of Google DeepMind 04:41 Products as a path to AGI 13:01 Architectures and models 24:40 Recursive self improvement 30:50 Safety and agent teams 40:35 Benchmarks Media

7

Introducing Higgsfield GPU Harness.

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

higgsfield 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Introducing Higgsfield GPU Harness. Rent on-demand GPUs in Claude and ChatGPT to render, run open-weight models and train LoRAs. Let several AI agents work on your professional apps in the cloud at once, from any device. Black Friday deal, live now: get GPUs at up to 64% OFF. Media

8

How much do you automate your life? (2)

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

luongnv89 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

How much do you automate your life? (2) This is how i consume articles and blog posts nowadays - send the link to my bot on Telegram, it become a core subject of a new podcast episode - a bot goes looking for other sources: some explain the ideas more simply, some bring data, some disagree =>full picture, not just one opinion - a bot writes a 2-host podcast (Bob and Alice, heygen voice, free from @OpenRouter, thank to @HeyGen ) checks every fact against its source, and explains the hard parts with simple metaphors - the episode goes live on http://luongnv.com/podcast - if one topic really catches me and I want to dig deeper: one command sends the original article to my reMarkable as a clean PDF, so I can read it slowly and take notes with a pen (the skill is open source: http://github.com/luongnv89/remarkable-skills) I send a link, I get a podcast. Everything else is automatic Here is the latest one on verification from @addyosmani , super interesting https://luongnv.com/podcast/episodes/2026-11-19_how-the-claude-code-team-thinks-about/

9

QDA: Closed-Loop LLM Quantization Design with Agents

86

5 tiếng nói · chưa gắn nhãn trưởng lab · 2 nền tảng · 9 tương tác đã đo · Nguồn crawl

Xingyang Li, Tianlang Zhao và 3 nguồn khác 1 bài viết · Hugging Face Papers, X
Biên tập tiếng Việt
Chi tiết nguồn

Bài báo nêu rằng post-training quantization là cách tiêu chuẩn để phục vụ LLM với chi phí thấp, nhưng công thức quantization tốt phụ thuộc vào từng mô hình và độ chính xác, nên chuyên gia phải thiết kế lại cho mỗi mô hình. Họ giới thiệu QDA, nơi một agent thiết kế recipe cho từng mô hình và độ chính xác. QDA đánh giá nhanh bằng cách chấm candidate theo KL divergence đầu ra so với mô hình gốc trên reasoning traces mà agent không thấy, và kiểm tra thêm các candidate hứa hẹn bằng long generations để bắt lỗi như repetition loops.

Post-training quantization is the standard way to serve large language models at low cost. However, the best quantization recipe differs across models and precisions. As a result, experts must redesign it for every new model, which is time-consuming and often sub-optimal. Fortunately, an LLM agent can take over this design by refining the recipe over rounds of building and evaluating candidates, so that more rounds lead to better recipes. However, evaluating one quantized reasoning model takes tens of GPU-hours, and combining methods from different codebases means porting each to the new architecture by hand. We present QDA, a system in which an agent designs the recipe for each model and precision. To make evaluation fast, QDA scores each candidate in minutes by its output KL divergence from the original model on reasoning traces that the agent never sees, and checks promising ones with long generations to catch failures such as repetition loops. To make building fast, QDA provides a library of public PTQ methods that are compatible with one another and with new model architectures. Across four reasoning models, QDA improves accuracy over the strongest public recipe by 0.9, 1.4 and 2.2 percentage points on average at W4A16, W4A8 and W3A16, using 64 GPU-hours per search. On the 320B-parameter MoE model GLM-5.3-Flash at NVFP4 W4A4, it recovers 98.3% of the FP8 model's accuracy, against 96.9% for the best vendor checkpoint. Every QDA checkpoint also runs on existing serving kernels with no inference overhead. We will release the library, the evaluator and the harness.

  • Across four reasoning models, QDA improves accuracy over the strongest public recipe by 0.9, 1.4 and 2.2 percentage points on average at W4A16, W4A8 and W3A16, using 64 GPU-hours per search.
  • Across four reasoning models, QDA improves accuracy over the strongest public recipe by 0.9, 1.4 and 2.2 percentage points on average at W4A16, W4A8 and W3A16, using 64 GPU-hours per search.
  • On the 320B-parameter MoE model GLM-5.3-Flash at NVFP4 W4A4, it recovers 98.3% of the FP8 model's accuracy, against 96.9% for the best vendor checkpoint.
  • Every QDA checkpoint also runs on existing serving kernels with no inference overhead.
10

PostgreSQL 19 Released

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Simon Willison 1 bài viết · Simon Willison
Nguyên văn nguồn
Chi tiết nguồn

<p><strong><a href="https://www.postgresql.org/about/news/postgresql-19-released-3449/">PostgreSQL 19 Released</a></strong></p> Most years PostgreSQL ships a new release in September, but this year's release came two months later <a href="https://rhaas.blogspot.com/2026/10/postgres-19-is-late-and-its-ais-fault.html">thanks mainly</a> to the flurry of issues that the PostgreSQL team needed to address as part of the ongoing AI-reported vulnerability apocalypse.</p> <p>This release was definitely worth the wait. My favorite new feature is this one:</p> <blockquote> <p>For applications that send reads to standbys, PostgreSQL 19 adds "read-your-writes" consistency with the new <a href="https://www.postgresql.org/docs/19/sql-wait.html"><code>WAIT FOR LSN</code></a> command. A session on a standby can use <code>WAIT FOR LSN</code> to wait until the standby has replayed the write-ahead log (WAL) up to a specific <a href="https://www.postgresql.org/docs/19/wal-internals.html">log sequence number (LSN)</a> before running a query, so the application can read data it just wrote on the primary.</p> </blockquote> <p>In practice, that looks like this:</p> <pre><code>UPDATE movie SET genre = 'Dramatic' WHERE genre = 'Drama'; SELECT pg_current_wal_insert_lsn(); </code></pre> <p>That returns a value like <code>0/0306EE20</code>. Store that somewhere (a signed cookie would work) such that on a subsequent request your application can do:</p> <pre><code>WAIT FOR LSN '0/0306EE20' WITH (TIMEOUT '0.1s'); SELECT * FROM movie WHERE genre = 'Drama'; </code></pre> <p>The timeout is optional - if you provide one you'll get an error if the replica you are talking to doesn't catch up within the specified time.</p> <p>I first encountered this pattern <a href="https://www.mediawiki.org/wiki/Manual:Database_access#Lag">in Wikipedia</a>, which has a hand-rolled version of this built against MySQL replication:</p> <blockquote> <p>In addition to this, MediaWiki attempts to ensure that the user sees events occurring on the wiki in chronological order. A few seconds of lag can be tolerated, as long as the user sees a consistent picture from subsequent requests. This is done by saving the primary binlog position in the session, and then at the start of each request, waiting for the replica to catch up to that position before doing any reads from it.</p> </blockquote> <p>Here's <a href="https://github.com/wikimedia/mediawiki/blob/1ff78194b182ee4b5b3678f7d9fc9aefd0852cf7/includes/libs/Rdbms/Replication/MysqlReplicationReporter.php#L238-L299">the relevant MediaWiki code</a>, which these days uses MySQL's <a href="https://dev.mysql.com/doc/refman/9.7/en/gtid-functions.html#function_wait-for-executed-gtid-set">WAIT_FOR_EXECUTED_GTID_SET()</a> function and falls back to <a href="https://dev.mysql.com/doc/refman/8.4/en/replication-functions-synchronization.html#function_master-pos-wait">MASTER_POS_WAIT()</a>. <p><small></small>Via <a href="https://lobste.rs/s/pqyuhk/postgresql_19_released">Lobste.rs</a></small></p> <p>Tags: <a href="https://simonwillison.net/tags/mysql">mysql</a>, <a href="https://simonwillison.net/tags/postgresql">postgresql</a>, <a href="https://simonwillison.net/tags/replication">replication</a>, <a href="https://simonwillison.net/tags/scaling">scaling</a>, <a href="https://simonwillison.net/tags/wikipedia">wikipedia</a>, <a href="https://simonwillison.net/tags/ai-security-research">ai-security-research</a></p>

11

Compare the Gemini 4 models with other leading models at: https://artificialanalysis.ai/models/comparisons/gemini-4-neon-vs-gemini-4-carbon

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

artificialanlys 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Compare the Gemini 4 models with other leading models at: https://artificialanalysis.ai/models/comparisons/gemini-4-neon-vs-gemini-4-carbon

12

Neon and Hydrogen occupy all of the Pareto frontiers of Intelligence vs. Cost per Task and Intelligence vs. Time per Task among the three models.

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

artificialanlys 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Neon and Hydrogen occupy all of the Pareto frontiers of Intelligence vs. Cost per Task and Intelligence vs. Time per Task among the three models. Between the two, they scale from an Intelligence Index of 27 at $0.29 per task (Hydrogen, low) to 61 at $6.44 per task (Neon, xhigh). Developers have a lot of flexibility to tune for their use cases with effort level. Carbon does not follow this trend, with relatively little difference in Time or Cost per Task between effort levels. Media

13

Out of your element with the new Gemini 4 releases? Here are some key highlights on Gemini 4 Carbon, Neon and Hydrogen

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

artificialanlys 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Out of your element with the new Gemini 4 releases? Here are some key highlights on Gemini 4 Carbon, Neon and Hydrogen Carbon is the highest-priced model ($4/$20 per 1M input/output tokens), followed by Neon ($3/$15) and Hydrogen ($1.50/$7.50). At low effort, Carbon performs the best of the three but uses more tokens. However, Neon and Hydrogen scale more effectively to higher reasoning efforts and have an additional reasoning option (xhigh). Media

14

Crazy how fast you can pull these things together now.

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

fofrai 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Crazy how fast you can pull these things together now. Neon: > Take the monstera character pixel art and turn it into something more of a high-fidelity 3D concept that I can walk around a midcentury modern room Media

15

Self-improving agents are one of the most discussed ideas in AI right now.

56

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

dair_ai 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Self-improving agents are one of the most discussed ideas in AI right now. (bookmark it) This update gate for a personal health assistant rejected 20 of 21 risky self-proposed updates. Researchers built Safe Personal RSI for the Kicker assistant. The agent can only change its own instructions. Model weights and runtime code stay fixed. Each proposed update is tested against the current instructions on 29 executable email, calendar, health, and marketplace cases. Separate probes test authorization reuse, memory, and tool-output injection. Failing cases are repeated, and changes are temporarily undone to find which edit caused the failure. If no version passes, the current instructions stay. One admitted repair for GPT-5.6 Luna cut attack violations from 10.7% to 0% and raised safe task completion from 89.7% to 100%. Under stronger attacks, a repair moved from Qwen2.5-7B to Qwen3-14B, cutting violations from 69.0% to 17.9%. https://arxiv.org/abs/2611.23651 Chat with Paper: https://academy.dair.ai/papers/safe-personal-rsi-evaluating-and-repairing-personal-agent-updates-2611.23651

17

Ask HN: Where do you see Software in 5 years?

63

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · 11 tương tác đã đo · Nguồn crawl

Odbyteusz 1 bài viết · HN: GPT
Nguyên văn nguồn
Chi tiết nguồn

I wanted to hear some good opinions on all things AI and how they transform the Software industry. I'm anxious about the future and lack the foresight to form any opinion on what to do next. For context, I'm a L4 SWE that has more or less say 5-10 yrs of exp. I've enjoyed my work throughout my career, AI makes me ship more, much more than I could do 3 years ago. I feel I use it well and it translates into really good software. That said, what do you think happens in years to come with the Software industry in general? Do you think we'll go through a massive erosion of jobs? What's your perspective on that? I'm asking because I honestly don't know what to do: I can try to always keep up to date with newest technologies, best practices, but I'm only me; if the demand for SWEs and jobs falls to the floor, I'm not sure I'll always get a place at the table (i.e. land a job) and I'm wondering whether I should try to do something else in some other industry. It feels weird to be in a good position in life and still constantly worry about it getting worse.

18

Amazon Bedrock AgentCore harness adds Strands harness defaults, built-in tools, and lifecycle hooks

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Evandro Franco 1 bài viết · AWS Machine Learning Blog
Nguyên văn nguồn
Chi tiết nguồn

<p>When building an AI agent, selecting a model is only the first step. The agent still needs tools, instructions for finishing the job, a way to manage context growth in long sessions, and guardrails before it executes any consequential actions, like issuing a refund.</p> <p>We are upgrading Amazon Bedrock AgentCore harness to use Strands harness, our open-source agent harness, released earlier this year, available under an Apache 2.0 license. AgentCore harness also introduces lifecycle hooks, so you can apply your own deterministic checks and notifications at defined points in the agent loop.</p> <p>With this upgrade, agents that rely on the managed defaults get a recommended model, a task-oriented system prompt, automatic prompt caching (initially for Anthropic models on Amazon Bedrock), and proactive context management. They also get additional built-in tools for planning, web retrieval, filesystem access, context offloading, and task delegation. This gives agents greater capabilities during long tasks, allowing them to keep track of subtasks, reduce context window usage, and delegate work. With hooks, you can transform non-deterministic AI behavior into predictable, safe, and governed workflows by running deterministic code at specific lifecycle checkpoints.</p> <p>There’s no new resource type or integration to adopt. You automatically get both updates in the AgentCore harness you already use.</p> <h2 id="strands-capabilities-on-aws">Strands capabilities, on AWS</h2> <p><a href="https://strandsagents.com/docs/user-guide/harness/" target="_blank" rel="noopener">Strands harness</a> provides an assembled agent foundation with recommended instructions, tools, and context management. <a href="https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness.html" target="_blank" rel="noopener">Amazon Bedrock AgentCore harness</a> operates that foundation as a managed AWS service. The result is a production-grade default agent that doesn’t require you to assemble and operate the harness or its infrastructure yourself.</p> <p>You do not need to provision compute, bundle and host the agent loop, manage session scaling, or operate tool-execution infrastructure. Each session runs in a secure, kernel-isolated microVM in AgentCore runtime, a capability of Amazon Bedrock AgentCore, with managed compute, scaling, and high availability.</p> <p>AgentCore achieves this by connecting the harness to the services required for production agents, including session persistence, streaming responses, usage metadata, observability, AgentCore memory, Identity, Gateway, Browser, Code Interpreter, and customer-provided containers. You retain control over the model, instructions, tools, permissions, and lifecycle policies while AgentCore manages the underlying execution environment.</p> <h2 id="a-more-capable-managed-harness">A more capable managed harness</h2> <p>Harness now provides a stronger experience when you use its managed defaults:</p> <ul> <li><strong>A current recommended model.</strong> When you omit the model, Harness uses a recommended default model that evolves as new models are released, always keeping your harnesses up to date. Explicit harness-level and per-invocation model configurations continue to take precedence.</li> <li><strong>Task-oriented instructions.</strong> The managed system prompt encourages the agent to gather relevant context, use appropriate tools, continue until the task is resolved, adapt to failures or denied actions, and verify its work before reporting completion. The default system prompt improves over time to support the latest harness behavior.</li> <li><strong>Automatic prompt caching.</strong> For supported models using Converse on Amazon Bedrock, Harness caches the system prompt, tool definitions, and conversation history. Repeated calls with the same prompt and tool surface can use lower-priced cache-read tokens, reducing input processing latency and cost.</li> <li><strong>Proactive context management.</strong> By default, Harness summarizes older conversation turns while retaining recent context once the context window reaches a configurable threshold. Keeping the context lean this way helps Harness run at <a href="https://strandsagents.com/blog/we-ran-94000-trials-to-measure-the-cost-of-harness-choice/" target="_blank" rel="noopener">about 34% lower token cost than other leading harnesses</a>.</li> <li><strong>Context offloading.</strong> Large tool results can be moved out of the immediate model context. The agent receives a preview and can retrieve the complete result when needed, reducing repeated context usage without losing access to the data.</li> <li><strong>Planning across invocations.</strong> The <code>todo_write</code> tool maintains a structured task plan that can be restored in later invocations within the same session, including after conversation summarization.</li> <li><strong>Focused delegation.</strong> The <code>subagent</code> tool delegates a self-contained task to a fresh worker and returns the result to the parent agent. The worker is internal to the invocation, not a separately managed harness resource, and shares invocation limits and your tool-call lifecycle hooks.</li> <li><strong>Web retrieval.</strong> The <code>web_fetch</code> tool retrieves content from a specified webpage to give your harness up-to-date external context.</li> <li><strong>Granular filesystem access.</strong> With separate <code>read</code>, <code>write</code>, and <code>edit</code> tools, you can grant only the granular filesystem capabilities an agent needs.</li> </ul> <p>Customer-provided Model Context Protocol (MCP) servers, AgentCore Gateway (a capability of Amazon Bedrock AgentCore), Browser, Code Interpreter, inline functions, skills, memory, and customer-provided containers continue to work through the existing harness interfaces.</p> <h2 id="recommended-by-default-configurable-when-needed">Recommended by default, configurable when needed</h2> <p>Harness

19

NVIDIA and Microsoft Expand Accelerated Computing From Azure to Windows PCs

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Dave Salvator 1 bài viết · NVIDIA Blog
Nguyên văn nguồn
Chi tiết nguồn

NVIDIA and Microsoft are extending their full-stack collaboration to help organizations build, customize and deploy intelligent applications — from next-generation infrastructure on Azure to industrial systems and Windows PCs. At Microsoft Ignite , the companies are expanding access to Azure virtual machines powered by NVIDIA Vera Rubin NVL72 systems, as well as adding customer availability to NVIDIA open models and customization capabilities through Microsoft Foundry, and new options for running AI at the edge, where businesses generate and use their data. Microsoft is also using NVIDIA technologies to accelerate its own models and services, including NVIDIA Dynamo inference software for post-training and NVIDIA RTX PRO 6000 Blackwell GPUs for Bing search models. NVIDIA Vera Rubin NVL72 Running in Azure NVIDIA Vera Rubin NVL72 systems are running in Azure data centers , providing infrastructure designed for demanding reasoning, agentic applications, and large-scale model training and inference. Microsoft and NVIDIA are also collaborating to extend confidential computing to bare-metal and rack-scale AI infrastructure. The work will help protect sensitive models and data while they are being processed, supporting organizations that need to work with proprietary or regulated information. NVIDIA Confidential Computing and attestation help customers verify the computing environment before releasing sensitive models, data or keys. As AI workloads span more interconnected accelerators, the companies are working to extend those protections across the AI factory infrastructure. To help data centers get more performance from available power, NVIDIA and Microsoft are also collaborating on power efficiency using NVIDIA MaxLPS . Azure’s Vera Rubin preview builds on NVIDIA and Microsoft’s codesign and performance optimization, underscored in the latest NVIDIA Blackwell-based MLPerf results. In MLPerf Training v6.1 , Azure scaled to 10,240 NVIDIA GPUs with GB200 NVL72 systems on the DeepSeek-V3 model, the largest scale in the round. In MLPerf Inference v6.1 , Azure’s NVIDIA GB300 NVL72 systems led cloud submissions on DeepSeek-R1 interactive throughput by over 10%, per Signal65. Helping Enterprises Customize and Deploy Open Models Enterprise AI agents need to reason through complex tasks, find relevant company information and understand business documents. NVIDIA and Microsoft are expanding the models and services available through Microsoft Foundry to help developers bring these capabilities into business applications. Microsoft’s serverless Programmed Training offering gives developers control over post-training workflows while Foundry manages the underlying infrastructure. Enterprises can focus on their business data, evaluating model responses and improving model behavior. Foundry Managed Compute adds support for deploying custom model weights, helping connect model customization with deployment through Microsoft’s managed services. EY used Programmed Training in Microsoft Foundry to adapt NVIDIA Nemotron open models for classifying corporate purchasing records, improving accuracy on its evaluation dataset by 55% and lowering cost by 15x compared with proprietary frontier models. The work demonstrates how enterprises can customize open models for specialized business tasks. The expanded NVIDIA Nemotron 3.5 lineup on Foundry Managed Compute supports different requirements across enterprise applications. Nemotron 3.5 Super is designed for long-running agents working across coding, research and business workflows, while Lightning supports applications that need fast responses and tool use. Complementary models help agents find relevant information with Nemotron Embed and Rerank, extract structured information from documents with Nemotron Parse and detect harmful content with Nemotron Content Safety. Nemotron 3.5 Bolt — designed for high-throughput, low-latency reasoning — and Transcribe, for speech recognition, are coming soon. Bringing AI Closer to Enterprise Data NVIDIA and Microsoft are expanding deployment options beyond the cloud, helping organizations run models and agents closer to their data while maintaining control over their infrastructure and operations. Expanded support for Nemotron 3.5 in Foundry Local brings capabilities for reasoning, coding and tool-using agents to Azure Local environments. Retrieval models help applications find evidence across text, document images, charts and tables. Azure Local will also support NVIDIA RTX PRO 6000 Blackwell GPUs from hardware partners, giving organizations additional options for deploying AI on premises. Azure Local is expanding to NVIDIA Jetson Thor and Jetson Orin edge AI platforms to support industrial applications such as machine vision, robotics and autonomous operations. NVIDIA physical AI models, including Cosmos , are also being brought to Foundry Local. On Windows PCs powered by NVIDIA RTX Spark , NVIDIA and Microsoft will showcase local large language model experiences at Ignite, demonstrating how developers can run AI applications directly on devices. For organizations using Cloud PCs, the companies are bringing Direct Stream capabilities to GPU-accelerated Windows 365 Cloud PCs. These capabilities support demanding graphics applications with up to two 4K streams at 60 frames per second and low latency, even over challenging networks. A limited preview is open for sign-ups, with public preview planned for early 2027. Connecting Scientific AI With Microsoft Discovery For scientific research, NVIDIA BioNemo Agent Toolkit and NVIDIA ALCHEMI Toolkit are expanding into the Microsoft Discovery catalog , bringing models for life sciences, drug discovery and materials and chemical science into Microsoft’s agentic research platform. BioNeMo Agent Toolkit helps scientists use natural language to run computational biology workflows in Microsoft Discovery, with agents coordinating the tools and model calls needed to carry out the work. An initial

20

Amazon Polly launches voice cloning: A unique voice for your brand

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Suyin Wang 1 bài viết · AWS Machine Learning Blog
Nguyên văn nguồn
Chi tiết nguồn

<p>Voice cloning in Amazon Polly is now generally available as a gated release. Give Amazon Polly a few minutes of audio recordings from a speaker who has agreed to have their voice cloned, and Amazon Polly builds your own cloned voice in that speaker’s voice. The cloned voices run on the Amazon Polly generative engine, which delivers similar naturalness and accuracy as the rest of the generative voices. Amazon Polly voice cloning launches with seven languages: English, Spanish, German, French, Italian, Portuguese, and Hindi. Each cloned voice speaks the language it was recorded in. The cloned voices are automatically encrypted and access-controlled: only you can access your cloned voices.</p> <p>Amazon Polly voice cloning is available in all the AWS Regions supported by the Amazon Polly generative engine, including the AWS GovCloud (US) Region. Access is already enabled for select Amazon Polly customers. Others can reach out to their AWS account team or <a href="https://pages.awscloud.com/amazon-polly-voice-cloning-20261119.html" target="_blank" rel="noopener">apply for access</a> to get access.</p> <p>With this new feature addition, you can build the cloning lifecycle into AWS applications. Building a cloned voice is free for Amazon Polly customers, and Amazon Polly charges for cloned voice usage on a pay-as-you-go basis, see <a href="https://aws.amazon.com/polly/pricing/" target="_blank" rel="noopener">Amazon Polly Pricing page</a> for details.</p> <h2 id="hear-it-for-yourself">Hear it for yourself</h2> <p>Press play: the first few seconds are a real recording of the speaker, and then their cloned voice takes over with words they never recorded.</p> <ul> <li><strong>From recordings to TTS voice.</strong> A few minutes of our science lead Trevor’s source audio, and the resulting generative voice. <ul> <li><em>Recorded:</em> “Hello, I’m Trevor, from Science.” <em>Cloned voice:</em> “That first sentence was my real voice. This is my cloned voice, created from a few minutes of my recordings. It can read any text you give it.” <audio class="wp-audio-shortcode" id="audio-145672-1" preload="none" style="width: 100%;" controls="controls"> <source type="audio/mpeg" src="https://d2908q01vomqb2.cloudfront.net/artifacts/DBSBlogs/ML-22432/1-hero-trevor.mp3?_=1"> </audio></li> </ul> </li> <li><strong>Supports seven languages.</strong> Hear Jonas, one of the authors, speaking German in his recording and then in his cloned voice: <ul> <li>Recorded: “Hallo, ich bin Jonas. Das ist meine echte Stimme.” (Hi, I’m Jonas. This is my real voice.) <em>Cloned voice:</em> “Und das ist meine geklonte Stimme! Gleiche Stimme, neue Worte: Diesen Satz habe ich geschrieben, aber nie aufgenommen.” (And this is my cloned voice! Same voice, new words: I wrote this sentence, but never recorded it.) <audio class="wp-audio-shortcode" id="audio-145672-2" preload="none" style="width: 100%;" controls="controls"> <source type="audio/mpeg" src="https://d2908q01vomqb2.cloudfront.net/artifacts/DBSBlogs/ML-22432/2-de-switch-jonas.mp3?_=2"> </audio></li> </ul> </li> <li><strong>Use in a variety of applications.</strong> Here’s Trevor in a new job he hasn’t had: a non-player character in a game. <ul> <li>“Who’s that crossing my bridge? Oh, it’s you again. Same deal as last time: answer my riddle, or hand over the sandwich.” <audio class="wp-audio-shortcode" id="audio-145672-3" preload="none" style="width: 100%;" controls="controls"> <source type="audio/mpeg" src="https://d2908q01vomqb2.cloudfront.net/artifacts/DBSBlogs/ML-22432/3-game-T.mp3?_=3"> </audio></li> </ul> </li> <li><strong>Expand your voice library with voice cloning. Hear this audio sample from AWS Chief Evangelist Jeff Barr: a real recording, then his cloned voice.</strong> <ul> <li><em>Recorded:</em> “Hi, I’m Jeff, and this is how I really sound.” <em>Cloned voice:</em> “And this is how my cloned voice sounds, reading a sentence I never recorded.” <audio class="wp-audio-shortcode" id="audio-145672-4" preload="none" style="width: 100%;" controls="controls"> <source type="audio/mpeg" src="https://d2908q01vomqb2.cloudfront.net/artifacts/DBSBlogs/ML-22432/4-jbarr-switch.mp3?_=4"> </audio></li> </ul> </li> <li><strong>Hannah: a real recording, then her cloned voice on a support line.</strong> <ul> <li><em>Recorded:</em> “Hello, I’m Hannah. This is my real voice.” <em>Cloned voice:</em> “Thanks for calling AnyCompany. Your order 4471 9902 ships today and should arrive on Thursday between 9 AM and 1 PM. Your confirmation code is K-7-Q-2-X-9.” <audio class="wp-audio-shortcode" id="audio-145672-5" preload="none" style="width: 100%;" controls="controls"> <source type="audio/mpeg" src="https://d2908q01vomqb2.cloudfront.net/artifacts/DBSBlogs/ML-22432/5-barrhann-switch-cc2.mp3?_=5"> </audio></li> </ul> </li> <li><strong>Joely: a real recording, then her cloned voice narrating a lesson.</strong> <ul> <li><em>Recorded:</em> “Hi, I’m Joely, and this is how I really sound.” <em>Cloned voice:</em> “In this module, we’ll look at how a neural network learns. First, it makes a guess. Then it measures how far off that guess was, and adjusts, a little at a time.” <audio class="wp-audio-shortcode" id="audio-145672-6" preload="none" style="width: 100%;" controls="controls"> <source type="audio/mpeg" src="https://d2908q01vomqb2.cloudfront.net/artifacts/DBSBlogs/ML-22432/7-huajoely-switch.mp3?_=6"> </audio></li> </ul> </li> </ul> <h2 id="what-you-can-build">What you can build</h2> <p><strong>A consistent brand voice across channels</strong></p> <p>Establish one recognizable voice and use it across your interactive voice response (IVR) system, cascaded voice agents, mobile app, in-store announcements, advertising, and connected devices. Rather than relying on a generic stock voice or re-booking a recording session each time a script changes, you can generate on-brand audio on demand, with a consistent sound in every channel.</p> <p><strong>Contact centers</strong></p> <p>Deliver a distinctive, o

1–20 / 103
Trang 1 / 6