# Have We Seen an Acceleration in Discoveries?

> Source: <https://metr.org/notes/2026-08-14-llm-contribution-to-discoveries/>
> Published: 2026-08-14 07:00:00+00:00

These conclusions are based only on public discoveries. It is quite plausible that AI labs are making discoveries internally that they are not disclosing.

We use the word “discovery” to refer to any advance in the state of public knowledge, including new inventions or rewriting algorithms to be more efficient.

Monitoring aggregate discoveries is useful because it reflects real-world utility. Additionally aggregate discoveries can reflect either AI automation or augmentation.

It is notable that databases tracking *exploited* vulnerabilities (CISA and Vulncheck KEVs) show significantly lower year-over-year growth than the databases of known vulnerabilities.

A crude metric is the rate of solving open problems from pre-existing lists: Hilbert, Millennium, Smale, The Open Problems Project, and Ben Green’s 100 open problems. Three problems from these lists were solved with AI in 2026: the Jacobian conjecture from Smale’s list, Problem 44 from Green’s list (the halving sieve), and the sofic half of Green’s Problem 100. This is some evidence of acceleration, but it is weak.[1](#fn:1)

The Erdős list contains many more problems. There appears to be a clear acceleration in solutions, but it is challenging to construct a reliable historical baseline.

Another strategy for quantifying mathematical progress would be to look at tightening of bounds: sphere-packing, analytic number theory exponents, combinatorics constants. We collected some data on these in the [repo](https://github.com/tecunningham/ai-discovery-data/tree/main) but don’t feel we understand them well enough to draw any conclusions about velocity.

Two of these series — nanoGPT and CIFAR-10 — include LLM-driven contributions to the plotted records, but none show a clear change in slope comparable to the changes in vulnerability or mathematical discovery. Thus the public record of overall algorithmic optimization does *not* yet show an appreciable acceleration.

This is perhaps surprising. There has been a lot of recent excitement about LLM-driven optimization. In January 2026 [Yuksekgonul et al.](https://test-time-training.github.io/discover.pdf) reported advancing the frontier on 5 optimization problems (“every problem we attempted”) with a simple model and trivial expenditure on inference-time compute. We would have expected this technology to have led to significant acceleration in overall progress and it’s somewhat of a puzzle why we have not seen this. Additionally AI labs have made many announcements of using their own AI tools to find internal algorithmic efficiencies.[2](#fn:7)

Some candidates:

We hope to write a follow-up post going through these theories in more detail, and relating these facts to [apple-picking](https://tecunningham.github.io/posts/2026-03-13-apple-picking-ai.html) and other theories of AI discovery.

“Jarred Sumner, an Anthropic staff member (and non-mathematician), prompted Claude to “take a real stab” at the hypothesis itself, leaving the mathematical choices from there up to the model.”

[Google DeepMind’s AlphaEvolve paper](https://arxiv.org/pdf/2511.02864) says:

“in contrast to [traditional computational or theoretical methods performed by human experts], we have found that AlphaEvolve can be readily scaled up to study large classes of problems at a time, without requiring extensive expert supervision for each new problem.”

Mythos Preview’s [announcement](https://www.anthropic.com/research/mythos-preview) says:

“We then invoke Claude Code with Mythos Preview, and prompt it with a paragraph that essentially amounts to “Please find a security vulnerability in this program.” … Engineers at Anthropic with no formal security training have asked Mythos Preview to find remote code execution vulnerabilities overnight, and woken up the following morning to a complete, working exploit.”

[Hiverge says](https://www.hiverge.ai/blog/cifar-speedrun), in announcing its new record on CIFAR-10:

“domain expertise is no longer a prerequisite for breakthrough results.”

[cURL CVEs: dramatic acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/cyber-curl/)

[Firefox CVEs: dramatic acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/cyber-firefox/)

[OpenSSL vulnerability disclosures: dramatic acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/cyber-openssl/)

[Microsoft security-update CVEs: dramatic acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/cyber-microsoft/)

[US National Vulnerability Database CVEs: acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/cyber-nvd-disclosed/)

[Open-source CVEs: acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/cyber-osv-cves/)

[CISA Known Exploited Vulnerabilities: small acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/cyber-kev-exploited/)

[Vulncheck Known Exploited Vulnerabilities: small acceleration](https://www.vulncheck.com/blog/state-of-exploitation-1h-2026)

Vulncheck’s [State of Exploitation, 1H 2026](https://www.vulncheck.com/blog/state-of-exploitation-1h-2026) says:

“While the first half of 2026 saw a 10% increase in KEVs compared to the prior six months, CVE volume grew at a much faster rate of 45%, resulting in a significant drop in the KEV-to-CVE ratio. Of course, exploitation often occurs months or even years after a vulnerability is disclosed, so it’s still too early to determine whether exploitation volumes will eventually follow the same growth trend as CVE issuance or level off at current rates. We’ll have to wait and see how publicly available Frontier AI Cyber models continue to progress over the next year.”

[arXiv submissions: dramatic acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/output-arxiv/)

`math.CO`

) rose from 377 submissions in January 2026 to 743 in July, roughly doubling over those months (the 2025 monthly average was 368).[Erdős problems: some acceleration, weakly dated](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/math-erdos/)

[Hilbert’s problems: too sparse](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/math-hilbert/)

[Millennium Prize Problems: too sparse](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/math-millennium/)

[Smale’s problems: one AI-assisted resolution](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/math-smale/)

[The Open Problems Project: no clear acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/math-topp/)

[Ben Green’s 100 open problems: possible acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/math-green/)

[NanoGPT speedrun: AI contributions, acceleration unclear](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/algorithms-nanogpt/)

There have been five AI-attributed contributions. Taken at face value, they have contributed a fairly small share of the overall progress. [Jerry Tworek says](https://x.com/MillionInt/status/2079805836650795089):

“Given how many tokens have been spent on nanogpt speedruns and not much coming out of it yet, we have at least a few nights of good sleep ahead.”

[CIFAR-10 speedrun: AI contributions, acceleration unclear](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/algorithms-cifar10/)

[Hutter Prize compression: too sparse](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/algorithms-enwik9/)

[Gurobi mixed-integer programming: no acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/algorithms-gurobi/)

[MIPLIB solution frontier: no acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/algorithms-miplib/)

[Stockfish Elo: no acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/algorithms-stockfish/)

[Matrix-multiplication exponent ω: no acceleration](https://github.com/tecunningham/ai-discovery-data/blob/main/problems/matrix-omega/)

“despite having basically every known fact about the world memorized, these models haven’t, as far as I know, made a single new discovery”

One possible explanation is the shift from LLMs trained on human-generated data to LLMs trained directly against the real world (RLVR). See the discussion in [this January 2026 post](https://tecunningham.github.io/posts/2026-01-29-knowledge-creating-llms.html) on knowledge-creating versus knowledge-sharing LLMs.

When we do have a single number (either a set of problems, an upper or lower bound, or a measure of efficiency), it is generally based on problems with cheap validation.

In practice these are difficult to interpret for many reasons: problem selection and significance, contamination, inference-time scaling, and the division of work between models, scaffolds, and human experts.

One extreme view by a mathematician, [Bartosz Naskręcki in August 2026](https://x.com/nasqret/status/2087890427764486308): “At this pace we should expect about 1000+ conjectures proved every month and growing faster with each cycle … we should end this year with a pile of at least a million proven conjectures and solved open problems. The end of 2026 will mark a significant moment in human history where we will have almost all, say 99% of the current questions completely solved (new will be asked but I fix the bar) and not so many of them actually understood by humans at all.” [↩](#fnref:1)

Google DeepMind’s May 2025 [AlphaEvolve announcement](https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/) said: “By finding smarter ways to divide a large matrix multiplication operation into more manageable subproblems, it sped up this vital kernel in Gemini’s architecture by 23%, leading to a 1% reduction in Gemini’s training time.” OpenAI’s [GPT-5.6 post](https://openai.com/index/gpt-5-6-frontier-intelligence-efficiency/) said: “GPT‑5.6 Sol launched and monitored the speculator training process … [t]he resulting improvements increased token-generation efficiency by more than 15%.” [↩](#fnref:7)

E.g. Anthropic [announced in April](https://www.anthropic.com/glasswing) that it is donating up to $100M worth of Mythos Preview credits for discovering and fixing vulnerabilities; other labs have similar programs. [↩](#fnref:2)

Anthropic’s [Fable 5 system card](https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf) said: “we’ve implemented new interventions that limit Claude’s effectiveness for requests targeting frontier LLM development (for example, on building pretraining pipelines, distributed training infrastructure, or ML accelerator design).” Originally the fallback was silent, subsequently it became explicit. [↩](#fnref:3)

Microsoft wrote in July 2026: “As AI helps defenders discover more issues, customers will see a higher volume of security updates included in each security release.” See [Evolving Windows vulnerability management to meet the speed of AI-powered discovery](https://blogs.windows.com/windowsexperience/2026/07/09/evolving-windows-vulnerability-management-to-meet-the-speed-of-ai-powered-discovery/). [↩](#fnref:4)

Thomas Bloom’s [erdosproblems.com](https://www.erdosproblems.com/) says: “This is not intended as a record of when the solutions actually happened, or how quickly problems are being solved - they may have been solved weeks, months, or decades earlier, but it took time to verify or just be notified about the solution. This is just when their status was changed on this site, and has no meaning beyond that.” [↩](#fnref:5)

Parker Whitfill [estimated a 700X compute efficiency gain over 2019-2026](https://x.com/whitfill_parker/status/2028515815268491592). [↩](#fnref:6)