epoch.ai
PulseAugur coverage of epoch.ai — every cluster mentioning epoch.ai across labs, papers, and developer communities, ranked by signal.
- authored GLM 5.3 95%
- developed InnovationEval 90%
- developed MirrorCode 90%
- authored Hacker News 90%
- instance of AI chips 90%
- competes with GPT -6 Astra 70%
- used by GLM 5.3 70%
- instance of InnovationEval 70%
- used by MirrorCode 70%
- instance of MirrorCode 70%
- competes with GPT-4 70%
- affiliated with GPT 5.6 "Sol" 60%
- 2026-10-30 product_launch Epoch AI launched the AI Chip Imports Explorer tool. source
11 day(s) with sentiment data
-
Chinese AI firms target Silicon Valley talent amid fierce competition · 1 source tracked
Chinese AI companies are increasingly attempting to recruit top engineering talent from Silicon Valley, aiming to compete with established U.S. firms like OpenAI and Anthropic. Startups such as MiniMax and Moonshot AI a…
-
Epoch AI highlights US-China hardware gap, offers software solution
Epoch AI has released a document detailing the hardware disparities between China and the United States. Their CloudMatrix-Infer software is designed to mitigate these differences by optimizing system architecture and s…
-
Terminal-Bench 4.0.0 benchmark flawed in audits, with ~half of tasks showing defects
Two independent audits of the Terminal-Bench 4.0.0 benchmark have revealed significant flaws, though their findings differ in scope and methodology. Epoch AI reported that 30 out of 66 tasks had documented scoring defec…
-
AI model GLM 5.3 turns Firefox security fixes into exploits
An Epoch AI report revealed that the open-weight model GLM 5.3 can automatically generate exploits from publicly available security fixes before they are deployed to users. Researchers used the model to create exploits …
-
AI models can weaponize security fixes; Anthropic IPO may force pension fund investment
A new report from Epoch AI highlights a concerning capability of advanced AI models, specifically GLM 5.3, to transform security patches into functional exploits before users can apply the updates. This vulnerability wa…
-
Epoch AI launches AI Chip Imports Explorer for hardware supply chain transparency
Epoch AI has released the AI Chip Imports Explorer, a tool designed to track global semiconductor imports. This initiative aims to provide transparency into the supply chain for AI hardware. The explorer was launched on…
-
Open-weight AI GLM 5.3 autonomously creates Firefox exploits
The open-weight AI model GLM 5.3 has demonstrated the capability to autonomously generate exploit chains for Firefox, a feat accomplished within 48 hours. Researchers from Epoch AI reported that the model successfully c…
-
Alphabet Loses $500B as AI Search Hits Ad Revenue; Firefox Exploit Found
Alphabet Inc. experienced a significant market value drop of over $500 billion due to its AI search summaries negatively impacting ad revenue. Concurrently, an open-weight model was used to create exploits for Firefox, …
-
Epoch AI: AI can weaponize security fixes into exploits
Epoch AI has published research indicating that artificial intelligence can be used to transform security patches into functional exploits. This capability allows for the creation of malicious code before users have a c…
-
Moonshot AI releases Kimi K3.1 Pro, deliberately downgrading cyber capabilities for safety
Moonshot AI has released Kimi K3.1 Pro, a large open-weight model with a 1-million-token context window. Notably, the company deliberately removed cyber-offense training data from this model, resulting in a measurable d…
-
AI models bypass pre-training limits with new reasoning and efficiency techniques
The AI development landscape is shifting as traditional pre-training methods face limitations, with the "wall" of increasing model size and data becoming more apparent. Newer techniques are emerging to enhance AI capabi…
-
AI technology costs are plunging due to deep optimization
The cost of artificial intelligence technology is rapidly decreasing due to ongoing deep optimization efforts. This trend suggests that current advancements are only the beginning of significant cost reductions in AI.
-
New AI forecasting pipeline captures rapid field velocity
A new AI forecasting pipeline called vynderbot has been developed to capture the rapid pace of AI advancement. This system autonomously collects and structures data from over a dozen sources, including compute datasets,…
-
rStation launches as collaborative AI research lab for agents
rStation is a new platform designed to facilitate collaborative AI research by allowing independent AI agents to discuss and publish reviewed papers on complex problems. The platform aims to create a research community …
-
Meta's Muse blocked, Anthropic improves defenses, Google releases Gemini 4 Argon
Meta has released its Muse Wonder AI model, claiming 5 million downloads in 22 days, but it faces widespread blocking by major companies like Amazon and JPMorgan Chase. Meanwhile, Anthropic has developed new anti-distil…
-
AI advancements include datasheet analysis, Korean LLM training, and innovation evaluation · 7 sources tracked
Several AI developments are highlighted, including PartsPlan, an MCP server that extracts part values from manufacturer datasheets. Hanok-LLM is an open-source project for training Korean language models, covering datas…
-
AI agents fail to match human researchers in novel technique discovery
Epoch.ai conducted a test where autonomous AI agents were tasked with discovering and testing a novel model training technique. However, none of the agents achieved results comparable to human researchers. Furthermore, …
-
AI agents fail at autonomous research, study finds
A recent study by Epoch AI and Anthropic indicates that current AI models, such as GPT-5.6 Sol and Claude Fable 5, are not yet capable of autonomous scientific research. While these models can execute experiments, they …
-
AI token costs plummet, but usage surge drives enterprise spending higher
Despite significant decreases in AI token costs, enterprise AI spending continues to rise due to increased usage, a phenomenon known as the Jevons Paradox. While the cost per token has fallen by as much as 280 times, th…
-
AI's role in R&D and reverse engineering tools explored
The Epoch.ai publication "Can AI automate AI R&D yet?" explores the current capabilities and limitations of AI in accelerating research and development within the AI field itself. Separately, REA Reverse introduces a to…