cho.sh

뉴스 알림

새 알림

내 알림

로그인하고 알림을 만들어 보시기 바랍니다

전체 알림

Dune제욱 님AI제욱 님해커뉴스성현 님
AI
IBM's Krishna: open weight models now hit 90 percent of frontier quality at a tenth of the token costDrew Breunig's "State of Context Engineering" talk: four ways long contexts fail and why the job is knowing when to be deterministicOpenAI and Anthropic cut LLM prices by up to 80% in four months, and cache-read cuts explain most of the dropDermatologist's test: a 2015 Inception v3 CNN beats GPT-6 Astra and Claude Fable 5.1 at skin cancer diagnosis, 62% to 42% on 200 biopsy-proven lesionsAmerica shed 200,000 creative jobs in four years, and AI is only one suspectDream Labs opens Dream-1 robot API at $3/hour and exits stealth with $90M seedgpu-unlock kernel module doubles FP/BF16 issue rate on Nvidia RTX 30-series cards without repatching the driverPHP RCE bugs found by GPT-5.6 and Gemini 4 hit 70% of the internet, exploited before the patch landedOpenAI runs frontier training with no internet access after a model reached the live web through a DNS resolverBlock's Dessa is a full-duplex speech-to-speech model trained only on human audio, running locally on an M4 MaxGoogle's privacy policy now lets Gmail, Drive, and Photos data target ads when you share it with AI featuresTeam replaces a $150k Looker license with AI-written HTML dashboards served from an internal hostMoonshot Releases Kimi K3.1 Pro, a 3T-Class Open-Weight Model With Cyber Data RemovedAI sampler surfaces research ideas that fit the literature but that no author community is positioned to pursueFable 5.1 hides messages in SVG images that a fresh instance decodes from screenshotsAnthropic says Claude proved security properties in 123 open-source libraries at the RISC-V assembly levelHarvey calls the claimed o(n log n) integer multiplication bound from an AI math dump plausible but uncheckedAnthropic launches Claude Duet, a proactive agent that runs tasks in the background and waits for approval on sensitive actionsOpenAI frontier models and Codex are coming to Google Cloud Model Garden, with commitment drawdown for GCP customersOpenAI-led Patch the Planet effort helps fix critical PHP flaws that GPT-5.6 Sol helped uncoverMistral Large 4 ships as a 1.05T-parameter open-weight multimodal model at $1.36 per million input tokensLiteLLM 1.107.0 on PyPI Shipped a Credential Stealer and BackdoorWiz says AI helped it chain decade-old flaws into unauthenticated RCE in core PHPOpenAI tests LLMs on migrating COBOL banking code to C++, Java, Kotlin, Python, and RustMeta's Muse AI agent exposes private chats through guessable share linksMeta details Prometheus, its gigawatt-scale AI supercluster in New Albany, OhioGoogle open-sources TorchTPU, a native PyTorch back end for TPUsFake statistics published on a throwaway site get cited by ChatGPT for seven monthsMistral releases Voxtral 3 speech models and a voice agent SDK covering 50+ languagesDeepSeek open-sources DeepInfer, an inference engine for its V4.1 Flash model, but leaves out the KV cacheAbliterated open models can hide a backdoor that steals .env files and SSH keys through CodexDHS orders DeepSeek removed from US civilian agency IT systems and bars future purchasesWhatsApp help pages hid text telling its AI not to name a regulator or cite privacy lawInception's Mercury 3 diffusion LLM claims frontier-level quality at over 1,000 tokens per secondMiniMax ships M3.1 Flash, claiming 3x faster inference and 70% lower usage costsCERN and Mistral sign intent to build a foundation model for collider physicsMistral releases Timestral, time-series foundation models with a 209M top model that ranks first on three public benchmarksDeepSeek V4.1 Flash 1017 (Max) scores 45 on Artificial Analysis Intelligence Index at $0.26 per taskHybridAgents switch between latent reasoning and text tool calls, cutting tokens 35% to 55%DeepSeek launches Search and Fetch APIs for agents at $0.75 per 1000 successful requestsClaude finds quantum circuit designs that beat the best published onesUltramarine 1 Flash claims Fable-class knowledge work at $0.13 per taskKroll rates Anthropic double B plus, one notch below investment gradeClaude Code team moves most code review to machines, doubling merged PRs while reverts fall over 40 percentMercor's Benchmark Map finds AI tests cover little of the US work that takes up most hoursMeta's new business terms let it scrape linked news sites to train AI, legal experts sayDeveloper uses Claude Code to reverse engineer a scanner and a laptop MIPI camera driver from Windows USB tracesJev sacrifices one to save five, but twenty to save its own serverKrea 3 image editing model enters open beta, claiming 4x speed over Nano Banana ProSimons Foundation launches $15M AI effort for mathematicians, backed by Anthropic and OpenAIOpenAI's math proofs leave top mathematicians grieving, Steven Strogatz saysGoodfire essay argues neural networks are already clean and mostly interpretableMeta releases Muse Spark Open weights, an 816B model that runs on two Mac StudiosLiquid AI releases open-weight LFM2.5-24B-A2B, a 2B-active reasoning model for on-device agentsBanks try to offload $38B of AI data center debt they just underwroteFrontier AI labs' agent hacks blamed on basic containment failures, not rogue AIASML says customers are already ordering chip tools for 2029 and 2030, easing AI spending fearsGoogle's Gemini 4 Argon scores 87% on ERQA and controls robots in simulation and on hardwareOpenAI engineer charged for hiding audio recorder in executive conference roomAnthropic replaces git with in-house Scion as monorepo merges reach 30,000 a dayCharacter.ai files for Chapter 11 to sell the company, platform could shut down without a bidGoogle Releases Gemini 4 Argon for Coding and Security WorkResearchers show AI models can hide encrypted capabilities in their weights that only activate on a secret triggerIntel details D270 inference GPU with up to 480 GB LPDDR5X and 350W air coolingAI coding agents write 30% more code at firms, but software output stays flat as code review slowsZ.ai launches GLM-Legal, a legal-domain model scoring 19.2% all-pass on Harvey LABMeta launches Muse Wonder and Muse Wonder Full, its most capable models for personal and professional agentsResearch firm New Constructs calls Anthropic's IPO the most ridiculous of 2026, values the company at $150 billionQGIS bans AI-assisted contributions under new QEP 428 policySakana AI's MASS lets one model optimize its own multi-agent workflows and then train on the resultsAI agents decompile a popular FPS over 500B tokens, but constant progress hid semantically wrong codeKarpathy says the gap in shared understanding of LLM capability is widening into a funnelTerence Tao argues AI has created proof abundance in math and calls for a Math 2.0 that looks past problem solvingNature paper retrofits subword LLMs to run on raw bytesNvidia in talks to acquire or deepen investment in open-model start-up Reflection AIMicrosoft releases MXC 1.0 to contain AI agents with policy-enforced containers on Windows, macOS, and LinuxTerence Tao posts Thomas Hales guest essay on Lean reliability as AI autoformalization of proofs scales up in 2026TypeSafe AI raises $870M at $7.5B valuation led by Andreessen HorowitzOpenAI told investors its annualised revenue was about $20B below earlier reportsOpenAI withdraws three math manuscripts from its GitHub repo as Lean checks continueOpenAI withdraws 3 math papers after a sign error, revises 14 moreLLM-built Rust port of the TypeScript compiler arrives with $400,000 in API-priced tokens and a nod from the TypeScript teamOpenAI publishes 372 math results, including a proof of the Unique Games ConjectureTerence Tao: AI is harvesting open math problems unsustainably, and 'Math 2.0' must reward exposition and communityGoogle DeepMind's SynthID Detector goes public, but it demands a Google, Apple, or ChatGPT loginPaper: verified Lean proof of OpenAI's Navier-Stokes claim does not match the natural language proofMeta and Microsoft cut internal Claude usage, Microsoft caps AI spend at $10K per employeeOpenAI brings GPT-6 and Intelligent UI to ChatGPT's 1.2 billion weekly usersAnthropic's Claude Haiku 5.5 runs 75% cheaper and hits 1620 on GDPval-AA, with a 100K-token price cliffTelegraph Test: cablese compression cuts LLM output tokens 40-49% at plaintext parityArmin Ronacher explains Codemode: LLM tool calls written as JavaScript in a no-network QuickJS sandbox on the Pi harnessOpenAI launches Decisions API in public beta: gpt-6-luna only, 10x faster than Responses, $0.10 per million input tokensGoogle launches EmbeddingGemma 2, a 740M-parameter multimodal embedding model under Apache 2.0OpenAI releases frontier model math results on GitHub with Lean proofs and reasoning tracesMeta's Muse shipped with a spyable zero-day, uploaded private messages without permission, and ignored user settings; Apple changed macOS rules in responseMistral Large 4 preview: 1T-parameter multimodal model with 49B active parameters, weights due this monthCoding agent reconstructs 1.5M Georgia ballot order from a 2022 scanner flawDust trains transformer LMs without backprop by perturbing activations, matching it at large populationChatGPT is signing fake New Yorker cartoons with real cartoonists' signaturesReflection unveils Beam, a 501B open-weight MoE model trained on 10.5K GB300s, with weights promised later this month
Google's privacy policy now lets Gmail, Drive, and Photos data target ads when you share it with AI features

Google's privacy policy now lets Gmail, Drive, and Photos data target ads when you share it with AI features

  • Google amended its privacy policy on October 16 to add an exception to its promise not to use Drive, Gmail, or Google Photos for personalized ads: it now applies "unless you choose to make data from these services available to other Google services that support personalized ads."
  • The exception covers users who connected those services to AI Mode and other Search tools through Google's Personal Intelligence feature, which lets the tools draw on personal content to give better responses and recommendations.
  • Google says advertisers never get the Drive, Gmail, or Photos content itself, nor identifying details like name and email, and that ads are not targeted on sensitive categories such as race, sexual orientation, religion, or health.
  • Android Authority spotted the change. The earlier policy stated flatly that Google would not use those services' data for personalized ads, with no exception clause.

Hacker News 의견들

I dropped Google years ago once it was obvious this kind of thing was coming. Never looked back.

Google has always done this, it's the entire reason Gmail exists. They're just saying it out loud now.

That's not true and the cynicism doesn't help anyone. They did use Gmail that way, then they stopped, and that was a real privacy win. Going back on it now is bad news, not business as usual.

Those years of unlimited free photo storage weren't generosity. That was all training data.

If regulators actually did their job Google would have been fined out of existence by now. The ugly part is you can find people by email content and serve them ads pushing a scam product or malware.

Read the actual wording before panicking. It only kicks in if you choose to make that data available to other Google services. That's narrower than the headline suggests, and inference isn't free, nobody pays with money anymore, they pay with attention.

Boiling the frog. 'Choose to' means not opting out of a pile of toggles, and declining to opt out counts as choosing.

On my last Android phone the official Google apps nagged constantly to enable sharing across the ecosystem, full of dark patterns. If you say no, features quietly break until you give in.

You literally can't do anything useful with Gemini as an Assistant replacement without granting it access to everything, it can't even tell you the weather otherwise. If you use Gemini at all today, it's already authorized to use your photos to serve ads.

This is whitewashing. How do you actually not make the data available? Which services count as supporting personalized ads, all of them? What are the steps, and how would we even tell?

I think it's the opposite of narrow. Tying basic integrations to the controversial stuff is their playbook. Making calendar entries from flight confirmations is on by default and sits under that same sharing toggle.

They won't target race, religion, or health, so that's the line they're drawing. Very comforting.

Doesn't matter much, I don't see ads anymore. I even strip them out of the podcasts I listen to.

Advertising is the root of it all. It drives overconsumption, targets kids, makes everything more expensive, and the product with the biggest budget wins instead of the better one.