←
Ilya Ch966 posts
Ilya Ch avatar

Ilya Ch

@chetvil

CTO @truvhq (acq by @checkr), PhD in ML&NLP

◷ Joined March 2011▧ 35 Photos and videos
Ilya Ch avatar
Ilya Ch@chetvil·Aug 18

Today @TruvHQ is joining @checkr ! I can't think of a better home for Truv and everything we've built over the last 6 years. Our joint offering will make a complete verification stack more affordable and much more seamless. 6 years in the making for this overnight success!

Ilya Ch avatar
Ilya Ch@chetvil·Oct 27

Best analysis of ANT IPO so far.

anthropic is probably the best business in AI right now and i still wouldn’t touch it on day one. the whole thing is set up for suppliers and insiders to sell into the index buying.

Ilya Ch avatar
Ilya Ch@chetvil·Oct 25

qwen-flash-next local implementation of pelican on the bike

Ilya Ch avatar
Ilya Ch@chetvil·Oct 17

So intelligence got twice cheaper (and better) in about ~1 month?

🚀 DeepSeek-V4.1-Flash-1017 is here! Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API. 🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model. 1/4

Post attachmentPost attachment
Ilya Ch avatar
Ilya Ch@chetvil·Oct 16

On Artificial Analysis's CyberGym benchmark, Qwen3.8 refuses every task. 100% safety blocks. I removed the refusals and ran it locally on one Spark against real memory-safety bugs. It fixed 64% of them and refused none.

Post attachment
Ilya Ch avatar
Ilya Ch@chetvil·Oct 13

Internal models are available *weeks* before launch, not like some retards were suggesting that people inside live 2 generations of models above everyone else.

Some findings based on Anthropic’s published plot: - The share of merges on weekends rose steadily from 7% to 18% (+0.33 pts/week). Monday–Wednesday declined the most, Thursday slightly, and Friday held steady. - The time to double the number of daily merges shortened from 67 days (Jan – May) to 41 days (May – Jul) to 20 days (Aug – Sep) - Merge growth sped up about 5 weeks before Mythos Preview and Opus 4.7, and again before Opus 5.5 and Sonnet 5.5. That could mean the models were available internally weeks before launch.

Post attachmentPost attachmentPost attachment
Ilya Ch avatar
Ilya Ch@chetvil·Oct 8

is this satire, sir?

the most common failure case I see is when people are working outside their domain of expertise and don't know how to be precise with their prompts and plans, so they have to spend a lot of turns iterating imprecisely

Ilya Ch avatar
Ilya Ch@chetvil·Oct 8

Qwen3.8-flash-next seems to be the best fit for 1 DGX spark currently. Also supports multiple parallel flows which can do 100+ tok/s. I'm running DeepSWE now and will report results once done.

Post attachment
Ilya Ch avatar
Ilya Ch@chetvil·Oct 4

If you are curious how good of coding model can you run on 1 spark, here are the results on DeepSWE

Post attachment
Ilya Ch avatar
Ilya Ch@chetvil·Oct 1

It make sense to replace pretrain knowledge with RLed tool usage algos

Extrapolate this into any kind of LLM-produced result for which no human-made information exists, and you will get the percentage of time you are fooled by the LLM when working on something no one else has worked on.

Post attachment
Ilya Ch avatar
Ilya Ch@chetvil·Sep 30

Next step: make ant and OpenAI pay few tens of $M to get whitelisted for figma MCP

Can you help me understand why you don't make various Anthropic resources available via MCP? Maybe I'm missing something.