NAll news

All news

Wednesday916 items

  1. AWS Machine LearningFirst-party
    AI score34

    How Amazon Quick integrates structured and unstructured data

    Amazon Quick uses the Space workspace to combine unstructured documents in an S3 knowledge base with structured datasets that have semantic descriptions, enabling a single conversational AI agent to automatically determine which type of data to query in a single question and return a merged answer with citations. This solution requires users to have an Enterprise subscription, the S3 bucket must be in the same region as Amazon Quick, and administrators can grant read-only permissions in the console.

    Architecture diagram showing the data, workspace, and agent layers that connect structured datasets and an Amazon S3 knowledge base to a chat agent in Amazon Quick

    #Tutorials / practice#RAG#Agents

  2. TechCrunch AI
    AI score64

    Anthropic Updates S-1: Quarterly Revenue $17 Billion, Compute Costs 58% of Revenue

    In an S-1 filing updated on Monday, Anthropic disclosed that revenue for the quarter ended September 30 reached $17 billion, up from $11.5 billion in the previous quarter, but compute costs rose from 48% to 58% of revenue, with operating losses widening in tandem. Compute spending for the quarter reached $10 billion, of which $5 billion came from short-term commitments, with costs significantly higher than comparable compute under multi-year contracts. The filing states that compute and infrastructure spending are expected to remain the largest operating expense for the foreseeable future, and that the company is building its own compute through the Anthropic Direct program, but demand growth means short-term neocloud contracts remain the primary choice.

    Same story · 21 reports

    #Industry news#English text is being prepared.#Anthropic

  3. Meta Engineering BlogFirst-party
    AI score36

    Meta F3 Lightning: How to Enable Independent Release of Ad ML Features and Achieve a 3.6x Speedup

    Meta introduces F3 Lightning, which builds on F3 by decoupling feature release from shared execution, enabling each feature to be independently compiled, validated, deployed, and rolled back, with similar features going live about 3.6 times faster than with the original F3 and without increasing infrastructure compute costs. F3 originally merged all features into a single graph as the release unit, so an error in a single definition would block the entire release process, while a single request now involves more than 20,000 features. Through four optimizations—stable computation identifiers, a single shared execution plan, cross-feature shared memory buffers, and dense index mapping—F3 Lightning achieves per-feature independent release while preserving the efficiency of shared computation.

    #Product updates#English text is being prepared.#English text is being prepared.

  4. Larsen Cundric
    AI score28

    Blogger comments on TikTok reverse engineering: about 25 models run on-device

    In response to Ayo O's post about reverse engineering TikTok, Larsen Cundric commented that running about 25 models on a phone all day without draining the battery is a fairly solid on-device infrastructure. He said that interpretations about surveillance are everywhere, but the engineering implementation behind it is astonishing.

    Quoted postAyo O · @ayoships

    I reverse engineered tiktok https://t.co/TEV1YKfAZQ

    #English text is being prepared.#English text is being prepared.

  5. Larsen Cundric
    AI score18

    Larsen Cundric Mocks the Airline Website Experience

    In response to Philip Kaplan saying that he used Grok @Bot to book a relatively complex family trip, had an excellent experience, and stated he might never use an airline's official website again, Larsen Cundric responded: No one has ever enjoyed an airline website.

    Quoted postPhilip Kaplan · @pud

    Used Grok @Bot to book flights for a relatively complex family trip. Incredible experience. I’ll likely never use an airline’s website again.

    #English text is being prepared.

  6. Ado Kukic
    AI score20

    Ado reposted the Claude Code October update recap

    ClaudeDevs recapped Claude Code's October updates: launched mods, improved the desktop app's ability to run multiple sessions simultaneously, and gave users a clearer understanding of where their usage goes. Ado's repost simply read "Shiptober!"

    Quoted postClaudeDevs · @ClaudeDevs

    October was a big month for Claude Code 🍁 We shipped mods, made the desktop app better at running lots of sessions side by side, gave you more insight into where your usage goes, and plenty more. Here's a recap! https://t.co/zZjzy654Uf

    #English text is being prepared.

  9. Garry Tan
    AI score12

    Garry Tan: Playing with air traffic control radar using AI bots

    Enhanced Radar announced the launch of the Enhanced Radar API, whose nationwide receiver network processes air traffic control audio in real time using the Y4 AI model. Garry Tan reposted and commented that it would be interesting for frequent flyers to play with this using muse, grok bot, Openclaw, or Hermes.

    Quoted postEric Button · @ericjacobbutton

    Today we're announcing the Enhanced Radar API. A few words on what this means. Background: at Enhanced Radar we operate a nationwide network of receivers to capture air traffic control audio. We process this data in real-time using our Y4 AI model. We use this data to... https://t.co/OV0NGMv6fM

    #English text is being prepared.

  10. Arena
    AI score41

    Tencent Hunyuan Hy4 is about to land on the Arena leaderboard

    Arena.ai has teased that Tencent Hunyuan's Hy4 is about to be released, and said it will be watching its ranking on the Arena leaderboard. The quoted post recalled that last month the Hy4 preview ranked 2nd in the Agent Arena open-source lab, ranked 2nd among open-source models (10th overall), with a net improvement of +5.4%, second only to Kimi K3 (Max)'s +6.6%; the cost per median task was $0.26, about 68% lower than Kimi K3 (Max)'s $0.80.

    Quoted postArena.ai · @arena

    Last month’s Hy4 preview launch landed @TencentHunyuan the #2 spot on the top 10 labs in Agent Arena among open-source labs! Across 14.5K+ real-world agent sessions, Hy4 preview ranks #2 among open models (#10 overall) with a +5.4% net improvement, trailing only Kimi K3 (Max) at +6.6%—a gap of just nearly 1 percentage point. This level of performance comes at roughly 68% lower cost: $0.26 per median task, compared with $0.80 for Kimi K3 (Max). Hy4 preview is now on the Agent Arena Pareto frontier. See the link below to explore its placement. By signal, Hy4 preview excels in: - Confirmed Success: 12% — explicit user feedback that the task worked - Praise vs. Complaint: 9% — implicit sentiment in user reactions - Bash Recovery: 7.5% — recovery from CLI errors It shows no issues with Tool Hallucination — calling tools that don’t exist By category among open models, it’s also #2 in Code and Work, and #3 in Chat. With this strong performance at a competitive price, Hy4 preview is a huge contribution to the open-source ecosystem. Congrats to the @TencentHunyuan team on this strong release!

    Same story · 2 reports

    #Model releases#Evaluations / benchmarks

  11. Thariq
    AI score20

    Thariq: The strongest models will no longer need to verify code

    Thariq commented on lauren's view that "adversarial code review is mostly a waste of tokens," saying this is basically the bitter lesson at work: the smartest models will need fewer and fewer tokens, and won't have to verify work or check for bugs—they will simply give the correct result directly.

    Quoted postlauren · @poteto

    maybe a hot take but i now consider adversarial reviews mostly a waste of tokens. far better to use those tokens on empirical combined with formal verification. i have updated pstack to use the interrogate skill less, and eventually i think i may remove it altogether i have found very little value with code review in the abstract, agents tend to just get lost in the nitpicks and yagnis

    #English text is being prepared.#Reasoning

  13. Peter YangProduct release
    AI score53

    Grok Bot Can Call US Businesses on Users' Behalf; Real-World Test Negotiates Lower Price and Faster Speeds with Comcast

    Peter Yang tested Grok Bot making a real call to Comcast customer service, successfully securing a lower price and faster internet speeds, saving nearly $500 a year. During the call, the bot answered the representative's question about the weather, declined 3 upsell attempts, and used a competitor's lower price as leverage. He said all Grok Bot users can now have the bot call US businesses, requiring only a phone number and background information.

    Quoted postPeter Yang · @petergyang

    Grok @Bot can now call US businesses for you, and it's one of my favorite AI features ever. Here's my new video where I play a real call of my bot successfully negotiating with Comcast to cut my internet bill by almost $500 a year. Some highlights include my bot: → Responding to "How's the weather?" → Turning down 3 upsells from the rep → Pushing the rep with a rival's lower offer 📌 Watch now: https://t.co/UWt3SanRSN

    Same story · 4 reports

    #English text is being prepared.#Agents#Speech

  14. Elon MuskProduct release
    AI score62

    Musk Says Grok Bot Can Research Nearby Doctors and Book Appointments by Phone

    Musk said on X that Grok Bot can research which doctors are best in the user's area and call to complete the appointment. The tweet cited Min Choi's firsthand experience: when his daughter injured her foot during dance practice, Grok Bot provided several nearby options, including a pediatric orthopedist with on-site X-ray capability, and called after they opened to book an appointment for 9:30 that same day; the entire call was transcribed in real time, and when the receptionist asked follow-up questions, it confirmed with him in real time, also obtaining insurance, payment method, and pricing information, sending the intake form to his email and adding it to his calendar; the receptionist knew the call was being recorded, and he could listen back afterward.

    Quoted postMin Choi · @minchoi

    Grok Bot just made a doctor's appointment for me while I was driving... My daughter hurt her foot landing a jump at dance practice. I told Grok Bot what happened, how she was feeling, and what I saw: not much swelling, iced, and her pain seemed manageable. It gave me a few nearby options, including a pediatric orthopedist with X-rays on site, and offered to call when they opened. Grok Bot called and got her in at 9:30 the same day. The whole call was transcribed live, and when the receptionist asked for more info, it asked me in real time. It also got the insurance, payment options and pricing, had the intake forms sent to my email, and put it on my calendar. The receptionist knew the call was recorded, and I could review the recording afterward. All while I was driving. Grok Bot is going to completely change people's lives.

    #Product updates#Agents#English text is being prepared.

  15. Rohan Paul
    AI score48

    Traversal raises $120M at a $1B valuation

    Rohan Paul commented that AI has greatly accelerated code writing, but running code in production has not become easier. Traversal fills in the missing production context for coding agents and feeds lessons learned from incidents back to these agents, enabling incidents to be traced back to the specific change or service that caused them. The company has just completed a $120 million funding round at a $1 billion valuation.

    Quoted postAnish Agarwal · @_anish_agarwal

    We just raised $120M, with a $1B valuation, to build software that keeps itself running. AI is writing more code than ever. It’s also making production less reliable: more code, shipped faster, understood by few. @traversal_ai understands how production systems fit together, finds the root cause of incidents, and helps prevent the next one. Engineers at Nubank, Harvey, PepsiCo, DigitalOcean, and a growing list of Fortune 100 companies use Traversal every day. Here’s production without the drama 👇

    Same story · 3 reports

    #Industry news#Agents#English text is being prepared.

  16. Rohan PaulProduct release
    AI score60

    Selendis launches video foundation model Archon, which can complete editing zero-shot and continuously learn according to creator preferences

    Selendis launches video foundation model Archon, which can zero-shot edit raw footage into a finished video and continuously learn each creator's editing preferences during use. Preference signals come from three sources: the creator's operational decisions in the Selendis multiplayer editor, their past video works, and the reference videos they cite. Selendis says that with every video edited, the editor becomes more accurate for that creator or team.

    Quoted postSelendis · @SelendisAI

    Introducing Archon, a video foundation model that zero-shots raw footage into finished edit in minutes. Built by Steve Aoki and industry experts for the next generation of content. Archon powers personal AI video editors that learn each creator’s taste from: - Decisions made in our multiplayer editor - Their past video catalogue - Reference videos they cite Every video edited makes the editor more accurate for that creator or team. Comment "SELENDIS" for early access + 3 months free. https://t.co/P2vkDbuLFm

    #English text is being prepared.#Video#Multimodal

  18. Claude DevsFirst-party
    AI score47

    Claude Code October update recap

    Anthropic's ClaudeDevs recaps the Claude Code October update: mods launched, the desktop app supports running more sessions in parallel, and users get a clearer picture of where usage goes. The thread also adds details such as idle sessions auto-compacting before the prompt cache expires, ultrathink raising effort for a single turn, subagent effort being specifiable, /usage explaining limit consumption, and timed-out shell commands moving to the background to keep running.

    #Product updates#Agents#English text is being prepared.

  19. Claude DevsFirst-party
    AI score42

    Claude Code idle session compaction reduces usage

    Claude Code will now compact long-idle sessions before the prompt cache expires (about 1 hour), so when you return to that session later, only the summary needs to be reprocessed rather than the entire conversation. When Pro and Max users resume very long conversations, if doing so would consume a large proportion of the 5-hour limit, they will receive a warning first.

    #Product updates#Agents#English text is being prepared.