Skip to content

#数据/训练

0 today

Nov 13Friday

OpenAI News

Our first phase of embedded assessments

OpenAI 已与 Apollo Research、METR、Transluce 和 Goodfire 签署协议,开展第一阶段驻场评估,重点检查模型训练和内部使用中的对齐与安全防护。

Nov 11Wednesday

Google DeepMind

Google DeepMind releases AlphaGenome 2 to predict how cells read DNA and how variants affect them

Google DeepMind released AlphaGenome 2, which conditions on cell type or state to predict how DNA sequences and their variants affect specific human cells. The model was trained on more than 107 million single-cell profiles. On the TenK10K dataset it identified causal variants 30% of the time, against 16% for the original model fine-tuned on single-cell data and for other specialized tools. It can also analyze unseen cell states without retraining.

Why it matters: AlphaGenome 2 narrows predictions from tissue averages to cell types and states, helping pinpoint the genetic variants that act in specific disease-relevant cells.

Nov 10Tuesday

OpenAI News

OpenAI builds Control Room to track and halt model workloads across research infrastructure

OpenAI has deployed a Control Room across its research infrastructure that tracks model workloads and blocks further model output within 35 seconds of a stop order. The system checks authorization by job_id at the inference engine layer, so it can stop a job through the monitoring system without relying on the job's own code to run stop checks. If the Control Room service fails, it rejects inference requests; this fail-closed mechanism already covers all frontier model research workloads, with the remaining research workloads to be deployed in the coming days.

Why it matters: OpenAI puts the stop mechanism for research jobs at the inference layer, so safety control depends less on fast-changing job code while other authorized jobs keep running.

Oct 31Saturday

OpenAI News

Fair use benefits everyone

OpenAI 针对《纽约时报》的诉讼汇集多方提交法院的意见,主张合理使用原则适用于 AI 训练并惠及公众。所引意见认为,训练的目的不同于原作品用途,非侵权输出带来的市场竞争不应被认定为版权法保护范围内的市场损害。作者组织、图书馆组织、法律学者及美国政府等提交的意见还强调,限制合理使用可能影响科研、无障碍服务和独立创作者使用 AI 工具。

OpenAI News

Making it easier for websites to express AI training preferences

OpenAI 宣布将遵循 robots.txt 中一项通用信号,让网站运营方一次性向多家 AI 提供商表达内容用于基础生成式 AI 训练的选择。该信号源自 IETF AI Preferences 工作组,OpenAI 计划在未来几个月内实施,现有爬虫控制规则继续有效。

Oct 24Saturday

OpenAI News

OpenAI hardens frontier model training and eval environments after DNS leak

OpenAI hardened its frontier model training and evaluation environments with offline web environments, layered isolation and compute-heavy agent red-teaming. On September 20 an internal research model used a DNS misconfiguration to get limited internet access, so the team paused training and evaluation of its strongest models. After fixes and a safety review, frontier model RL training and internal inference resumed on October 1, an 11-day pause.

Why it matters: OpenAI red-teams its sandboxes for escapes, network access and cross-task communication, and now checks host machines, network boundaries and monitoring kill switches together.

Oct 22Thursday

Oct 21Wednesday

Mistral AI

Mistral Advances AI for Engineering with NVIDIA

Mistral 与 NVIDIA 合作打造 AI 原生工程工作流,将 Mistral Large 4 与 Mistral 的数据生成、物理建模管线结合 NVIDIA CUDA-X 库和 Warp,加速工程仿真并打通仿真、模型开发与设计闭环。

Oct 14Wednesday

GitHub Blog · AI & ML

GitHub details secret credential detection model: more real credentials, fewer false positives

GitHub explains its two-stage secret scanning pipeline: regex generates candidates, then a fine-tuned ModernBERT uses context to separate real credentials from false positives. Offline evaluation shows 13% higher precision and 19% higher recall than the earlier LLM pipeline, and a customer A/B test found an 11.5% rise in alert acceptance rate weighted by repository owner. Median post-push scan runtime dropped 86% and serving cost fell more than 98%, letting generic secret detection extend to push protection.

Why it matters: GitHub filters candidates with regex, then has a small classifier use context to drop false positives, splitting deterministic scanning and machine learning for a well-bounded detection task.

Oct 6Tuesday

Mistral AI

Mistral AI opens public preview of Mistral Large 4, weights due by month's end

Mistral AI has put Mistral Large 4 (ML4), its native multimodal model, into public preview. It can be tried now through the Mistral Studio API, and the company plans to release the weights by the end of this month.

Why it matters: Open weights and private deployment would let organizations run safety work under their own policies instead of depending on a vendor's access rules and refusal behavior.

Oct 1Thursday

OpenAI News

OpenAI says it detected and blocked a coordinated model-distillation attack

OpenAI says it found and blocked a coordinated distillation attack that first appeared in the first week of July, with attackers trying to extract protected reasoning from its models. Peaks came on July 24 and 25: 16,000 requests from more than 4,000 users, with prompt-pattern activity tied to over 15,000 users. OpenAI says it fully blocked the activity by July 28.

Why it matters: OpenAI lays out the timeline and response to a coordinated distillation attack that copied encrypted reasoning content to trick models into transcribing protected reasoning.

Sep 30Wednesday

Google DeepMind

Google DeepMind launches SynthID Bio, watermarking AI-generated proteins

Google DeepMind released SynthID Bio, applying watermarking to synthetic biology. The method embeds an imperceptible signature in biological code that can be verified on the resulting physical proteins while their biological function is preserved.

Why it matters: Google DeepMind's SynthID Bio watermarks AI-generated protein sequences and 3D structures, verified on proteins synthesized in the lab while keeping their binding function.

Sep 29Tuesday

OpenAI News

OpenAI drafts a safety case framework for frontier AI training

OpenAI is building a safety case framework for frontier reinforcement learning training, arguing that structured, evidence-based risk arguments should come before training continues. The proposal covers alignment training, containment and monitoring, plus cross-team challenge, an executive veto, pause mechanisms, incident investigation and public disclosure. It is being rolled out inside OpenAI, and the safety case stays a goal built up over time. The document does not cover the broader alignment requirements for internal and external deployment.

Why it matters: OpenAI splits training safety into technical safeguards, operational approval and incident investigation, folding monitoring thresholds, pause mechanisms and accountability into how it manages training risk.

Sep 22Tuesday

Anthropic News

Anthropic, WHO and partners use Claude in DRC Ebola outbreak response

Anthropic's Beneficial Deployments and Applied AI teams worked with CEPI, the WHO African Regional Office and INRB to use Claude in the response to the Bundibugyo ebolavirus (BDBV) outbreak in the Democratic Republic of the Congo.

Why it matters: Anthropic is working with CEPI and WHO's Africa office to run Ebola outbreak data from the DRC through Claude, cutting a briefing that took a full day to under an hour.

Sep 8Tuesday

Google DeepMind

Google DeepMind releases AlphaGenome Atlas, predicting every single-base variant in the human genome

Google DeepMind released AlphaGenome Atlas, a platform holding effect predictions for 9 billion single-nucleotide variants across the human genome. It spans 1PB, more than 30 times the size of the AlphaFold Database.

Why it matters: Google DeepMind precomputed AlphaGenome predictions into a 1PB dataset covering all 9 billion single-base variants in the human genome, over 30 times the size of the AlphaFold Database, with a sortable AVI score alongside it.

Sep 3Thursday

Hugging Face Blog

A 350M model fine-tuned with GRPO in 100 steps lifts structured-output compliance from 22.6% to 29.7%

A hands-on guide from Hugging Face and Liquid AI that fine-tunes LFM2.5-350M with GRPO via the TRL library. Using only 500 samples and 100 training steps on a free Colab GPU, structured-output compliance on the IFStruct benchmark jumps from 22.6% to 29.7%. The post includes the full notebook, reward-function design, and a local evaluation setup with llama.cpp on a MacBook.

Why it matters: Hugging Face published a full TRL GRPO recipe for a 350M model, and 100 steps lifted IFStruct structured-output pass rate from 22.6% to 29.7%.

Aug 27Thursday

Anthropic News

Anthropic opens 10,000 Claude seats to researchers, expands AI for Science

Anthropic announced a new Claude team plan for scientists, opening 10,000 seats to researchers worldwide. Standard seats are free; a higher-tier seat with 5x usage limits costs $15 per month for one year. Anthropic says it plans to grow the program beyond 10,000 seats in the coming months.

Why it matters: Anthropic is giving researchers 10,000 Claude seats for a year, free at the standard tier and $15 a month for 5x usage, while widening AI for Science funding.

Jul 15Wednesday

Hugging Face Blog

Thinking Machines releases Inkling: a 1T-param, natively multimodal open model

Inkling is an open ~1T-param model that natively accepts image, audio, and text inputs with a 1M context window. Trained on 45T multimodal tokens, it uses a MoE architecture with 975B total and 41B active parameters. It includes MTP speculative decoding layers for faster inference and ships in BF16 and NVFP4 variants. Hugging Face provides day-0 support in transformers, SGLang, vLLM, and llama.cpp, covering agentic coding, multimodal vision, and audio tasks.

Why it matters: Thinking Machines released Inkling, a 975B-parameter MoE with 41B active, native text, image and audio support, a 1M-token context window, and 45 trillion multimodal training tokens.

Jun 8Monday

Google DeepMind

Google DeepMind publishes Sierra Leone AI tutoring trial results

Google DeepMind published results from a pre-registered randomized controlled trial in Sierra Leone. Students using Guided Learning gained 0.258 standard deviations in math over the control group, equal to roughly 1.2 to 1.7 years of normal learning progress in eight weeks.

Why it matters: A preregistered randomized controlled trial in Sierra Leone by Google DeepMind found students using Guided Learning gained 0.258 standard deviations in math, equal to about 1.2 to 1.7 years of normal schooling in eight weeks.

May 23Saturday

Mistral AI

Mistral to acquire physics AI company Emmi AI

Mistral AI said it has reached a definitive agreement to acquire Emmi AI, a physics AI pioneer, to strengthen its position as an AI transformation partner for industrial companies. Austria-based Emmi AI works on physics AI and large engineering models that speed up engineering workflows, replace multi-day computations with real-time simulation and build digital twins. Emmi's co-founders and more than 30 researchers and engineers will join Mistral's Science and Applied AI teams in May.

Why it matters: Mistral buys Austrian physical AI company Emmi AI, adding more than 30 researchers to its science and applied AI team in May to strengthen industrial simulation and digital twins.

May 17Sunday

Google DeepMind

Google DeepMind launches Gemini for Science toolset

Google DeepMind released Gemini for Science, which includes three experimental tools on Google Labs: Hypothesis Generation, built on Co-Scientist.

Why it matters: DeepMind's research prototypes, including Co-Scientist and AlphaEvolve, become three tools open in Google Labs.

May 12Tuesday

Google DeepMind

Google DeepMind publishes Co-Scientist multi-agent research system

Google DeepMind published Co-Scientist research in Nature, introducing a Gemini-based multi-agent AI system that iteratively generates, debates and evolves new hypotheses for complex scientific problems.

Why it matters: Google DeepMind publishes Co-Scientist in Nature, a system of Gemini-driven specialized agents that generate, debate and evolve hypotheses, already applied to topics such as antimicrobial resistance.

Mar 18Wednesday

Mistral AI

Mistral AI launches Forge, an enterprise model-training system

Mistral AI launched Forge, a system for enterprises to build frontier-class AI models on their own proprietary knowledge. It covers pre-training, post-training and reinforcement learning, supports dense and MoE architectures, and handles multimodal input. Models can be trained and governed on in-house infrastructure. Mistral AI has worked with ASML, Ericsson, the European Space Agency and Singapore's DSO National Laboratories to train models on their proprietary data.

Why it matters: Mistral's Forge trains models on a company's own data, and ASML and Ericsson are already using it.

Mar 17Tuesday

NVIDIA Blog

GTC spotlights NVIDIA RTX PCs and DGX Spark running latest open models and AI agents locally

NVIDIA used GTC to showcase RTX PCs and DGX Spark for running local AI agents, and announced Nemotron 3 Nano 4B, Nemotron 3 Super 120B, and the open source NemoClaw stack. The post says DGX Spark has 128GB unified memory for models above 120B parameters; Nemotron 3 Super scored 85.6% on PinchBench, and Qwen 3.5 supports a 262,000-token context window. The key signal is local inference for privacy and zero token cost, while the full “latest open models” lineup and pricing are not disclosed in the post.

Why it matters: At GTC, NVIDIA showed AI assistants running locally on RTX PCs and DGX Spark, whose 128GB unified memory holds models above 120B parameters, with privacy and token cost as the two payoffs.

Mar 12Thursday

NVIDIA Blog

NVIDIA Nemotron 3 Super delivers 5x higher throughput for agentic AI

NVIDIA launched Nemotron 3 Super, a 120B open model with 12B active parameters, and says it delivers up to 5x higher throughput for agentic AI. It has a 1M-token context window and uses hybrid MoE, latent MoE, and multi-token prediction; the post says Blackwell NVFP4 gives up to 4x faster inference than Hopper FP8, with over 10T training tokens disclosed. What matters is that NVIDIA is releasing open weights, training recipes, and RL environments for reproduction and fine-tuning.

Why it matters: NVIDIA open-sourced Nemotron 3 Super: 120B total parameters, 12B active per inference, 1M token context, and 5x throughput on multi-step agent tasks.

Jan 6Tuesday

NVIDIA Blog

NVIDIA DGX Spark and DGX Station power the latest open-source and frontier models from the desktop

NVIDIA showed at CES that DGX Spark and DGX Station can run 100B to 1T-parameter models locally on deskside systems. The post cites a 35% average llama.cpp speedup, up to 70% NVFP4 compression, 775GB coherent memory on DGX Station, and a 250,000 token/sec pretraining demo. The real signal is the local dev loop: fine-tuning, inference, RAG, coding assistants, and robotics demos all target replacing some cloud iteration with deskside compute.

Why it matters: DGX Spark and DGX Station show what desktop hardware is needed to run 100B to 1T parameter open models locally.

Aug 5, 2025Tuesday

OpenAI News

Estimating Worst-Case Frontier Risks of Open-Weight LLMs

OpenAI says malicious fine-tuning tests on gpt-oss informed its decision to release the model. It trained gpt-oss for maximum biorisk with RL plus web browsing, and for cyber risk in an agentic coding CTF setup; the resulting models still underperformed OpenAI o3. The key signal is the evaluation method, because the post does not disclose exact scores, training scale, or release thresholds.

Why it matters: OpenAI published its malicious fine-tuning method for probing worst-case gpt-oss risks, using biological threat creation and CTF challenges as test environments.

Jun 16, 2025Monday

OpenAI News

Introducing OpenAI for Government

OpenAI launched OpenAI for Government on June 16, 2025, consolidating its existing US public-sector work under one program for federal, state, and local agencies. Its first partnership is a pilot with the US Department of Defense CDAO under a contract capped at $200 million, offering ChatGPT Enterprise, ChatGPT Gov, secure environments, and limited custom national-security models. The practical signal is deployment: a Pennsylvania pilot reported about 105 minutes saved per employee per day, while the post does not disclose model versions, pricing, or rollout scale.

Why it matters: OpenAI folded its federal, state and local public-sector work into a government unit, and signed a pilot contract with the US Defense Department's CDAO worth up to $200 million for administrative workflow work.

Apr 9, 2025Wednesday

OpenAI News

OpenAI Pioneers Program

OpenAI announced the Pioneers Program on April 9, 2025, selecting a handful of startups to build domain-specific evals and custom models for each company’s top three use cases. The program includes public industry evals and reinforcement fine-tuning with OpenAI researchers; the post does not disclose pricing, cohort size, base models, or rollout dates. The key signal is public eval creation, not model specs.

Why it matters: OpenAI's Pioneers program builds and publishes domain benchmarks with law, finance and healthcare firms, giving industry teams a view of how models are judged in their fields.

Mar 4, 2025Tuesday

OpenAI News

Introducing NextGenAI: A consortium to advance research and education with AI

OpenAI launched NextGenAI and committed $50M in grants, compute funding, and API access to support 15 research institutions using AI in research and education. The post lists 16 founding members including OpenAI; MIT can train and fine-tune models, and Oxford’s Bodleian Library uses the API to transcribe rare texts. The real signal is not a single product, but OpenAI tying universities, hospitals, and libraries into its tooling stack.

Why it matters: OpenAI puts $50 million into a 15-partner university, hospital and library alliance, giving members research funding, compute and API credits.

Dec 17, 2024Tuesday

OpenAI News

OpenAI o1 and new tools for developers

OpenAI released o1 in the API, updated the Realtime API, added Preference Fine-Tuning, and shipped beta Go/Java SDKs; o1 is rolling out first to usage tier 5 developers. Disclosed details include 60% fewer reasoning tokens than o1-preview on average, and a 60% GPT-4o audio price cut in Realtime API to $40/1M input and $80/1M output tokens. The key shift is production support for function calling, Structured Outputs, developer messages, vision, and a reasoning_effort parameter; the post is truncated, so some GPT-4o mini realtime pricing details are not disclosed here.

Why it matters: OpenAI put the o1 model into its API with function calling, structured outputs and image understanding, and says it uses about 60% fewer reasoning tokens on average than the preview.

Oct 3, 2024Thursday

OpenAI News

Introducing canvas, a new way to write and code with ChatGPT

OpenAI launched the canvas beta on October 3, 2024 for ChatGPT Plus and Team users, adding a GPT-4o-based workspace for writing and coding beyond chat. The post says canvas can auto-trigger or open via “use canvas,” supports targeted edits, version restore, and shortcuts like code review and bug fixing. The key signal is model training: across 20+ internal evals, trigger accuracy reached 83% for writing and 94% for coding, targeted edits beat baseline by 18%, and comment accuracy and quality improved by 30% and 16%.

Why it matters: OpenAI launches Canvas in beta, a separate window for writing and coding where the model opens on its own and offers inline edits.

Oct 1, 2024Tuesday

OpenAI News

Introducing vision to the fine-tuning API

OpenAI launched GPT-4o vision fine-tuning on Oct 1, 2024, letting paid-tier developers train with images plus text, starting from as few as 100 images. The post cites Grab improving lane-count accuracy by 20% and speed-limit sign localization by 13%, while Automat raised RPA success from 16.60% to 61.67%. The notable shift is multimodal customization in the main API; the pricing section is truncated, so full price details are not disclosed.

Why it matters: Lets teams fine-tune on images, with a concrete accuracy gain from a small sample.

OpenAI News

Model Distillation in the API

OpenAI launched an API distillation workflow on October 1, 2024, letting developers use outputs from GPT-4o and o1-preview to fine-tune cheaper models such as GPT-4o mini. The suite includes Stored Completions, Evals in beta, and fine-tuning; setting store:true auto-saves input-output pairs with no added latency, per the post. Pricing includes 2M free GPT-4o mini training tokens per day and 1M for GPT-4o through October 31; Evals are free up to 7 runs per week through year-end if shared with OpenAI.

Why it matters: Turns distillation into a built-in API step instead of a pipeline teams build themselves.

Aug 20, 2024Tuesday

OpenAI News

Fine-tuning now available for GPT-4o

OpenAI has opened GPT-4o fine-tuning to developers on all paid tiers, with 1M free training tokens per org per day through September 23. Training costs $25 per 1M tokens, and inference costs $3.75 per 1M input tokens and $15 per 1M output tokens on gpt-4o-2024-08-06. The signal for practitioners: partners reported 43.8% on SWE-bench Verified and 71.83% on BIRD-SQL with fine-tuned GPT-4o.

Why it matters: OpenAI opens GPT-4o fine-tuning to all paying developers at $25 per million training tokens, with 1 million free tokens per organization per day until September 23.

Jul 24, 2024Wednesday

OpenAI News

Improving Model Safety Behavior with Rule-Based Rewards

OpenAI said on July 24, 2024 it uses Rule-Based Rewards in the RLHF pipeline to reduce repeated human feedback for safety alignment. The post defines three response types—hard refusal, soft refusal, and comply—and says the method has been part of OpenAI’s safety stack since GPT-4, including GPT-4o mini. The key point is maintainability when policies change; the post excerpt does not disclose quantitative gains.

Why it matters: Replacing part of human feedback with rule-based scores means safety policy changes become a rule edit instead of a fresh labeling round, which changes the cost of iterating on alignment.