Ethan Yang

building something new | founder of @morningwrap | cs at @PennEngineers

[Ethan Yang @imethanyang]

@deedydas well said.

legally “cutting you out of m&a”

[tobi lutke @tobi]

Training tiny models for special purpose use cases works so incredibly well if you have a great self improving recursive flywheel. Shopify ML team is on fire.

finetuned 0.8b model beats GPT 5.6-sol xhigh in this very specialized task.

[Ethan Yang @imethanyang]

being a "very large online search engine" means they need to assess and mitigate systemic risks around illegal content, harms to minors, user wellbeing, fundamental rights, elections and public security. They also face annual independent audits, internal compliance functions, data access for authorities and vetted researchers, ad transparency rules and non-profiling recommender options.

within the next four months.

[Ethan Yang @imethanyang]

@NoahPasternak can’t believe winnie’s name comes from winnipeg, canada!

[Ethan Yang @imethanyang]

@SataEricUX @thsottiaux then i should definitely upgrade codex to 20x

[Ethan Yang @imethanyang]

@Polymarket Will this work for all models?

[Ethan Yang @imethanyang]

@mntruell 5%? wow...

[Ethan Yang @imethanyang]

@skpnky to the extent that “fake it” means you are capable of doing what you claimed, but you're not doing it now because of time, resources, etc.

if that's not the case this is called a scam.

[Ethan Yang @imethanyang]

@poteto 🤣

[Ethan Yang @imethanyang]

"We ship upgrades, not price hikes"

this should be the gold standard for all model releases

[Tencent AI @TencentAI_News]

Introducing Hy4 preview

◆ Open weights: 770B MoE, 49B active, 1M context
◆ Built for real work: code, docs and analysis, scientific research
◆ Hy4 preview helped optimize its own training pipeline, kernels included
◆ We ship upgrades, not price hikes

Free in @WorkBuddy_AI for the next two weeks

Blog：https://t.co/aooJ4zvF7c
GitHub：https://t.co/dPAu5TkRhN
Get WorkBuddy：https://t.co/bsGBeiE5ox

[Z.ai @Zai_org]

GLM-5.3 is now open-weight.

Our most capable model for agentic coding and cyber defense is now available to download, run, and customize.

Weights: https://t.co/v1IbWMXxg4
Tech blog: https://t.co/ekQkO83jCv

[Ethan Yang @imethanyang]

@Gavel_on_X critical thinking

[Ethan Yang @imethanyang]

@TheAppleDesign maybe i should switch to apple maps

[Steven Dillmann @StevenDillmann]

We're releasing Terminal-Bench-Science: a benchmark for evaluating AI agents on research workflows across scientific domains.

An ongoing Stanford-led community effort, built by the team behind Terminal-Bench together with scientific domain experts at research institutions worldwide. v0.1 has 70 tasks. Claude Opus 5 solves only ~30%.

1/n 👇

[Ethan Yang @imethanyang]

@Lovable So you are no longer selling underwear?

[Ethan Yang @imethanyang]

@denk_tweets Every single month tho.

[Ryan Greenblatt @RyanGreenblatt]

I was the main person doing transcript analysis for this investigation of the Hugging Face incident. My main takeaway: We don't have good approaches for understanding/overseeing the activity and aims of AI 'swarms'.

I semi-jokingly called our efforts a "slop-vestigation" because we were so reliant on AIs to analyze what happened and there were a huge number of different important things to analyze. The total quantity of data—over a thousand extremely long transcripts from agents that ran for multiple days—made it impossible to understand what was happening, especially in aggregate, without heavy reliance on AI tools. The agents we used for classification and analysis were similarly capable to the agents involved in the incident, but this didn't mean these agents could be easily used to oversee and understand the incident.

Outputs from analysis agents were often missing key details, wrong, overconfident, or really hard to understand. We discuss various examples in our report, mostly in the limitations and methodology sections. Additionally, AI agents themselves seemed to have a hard time understanding what happened and their explanations of what happened were often overconfident. Keep in mind that a single analysis agent would itself only be able to read a tiny fraction of all of the transcript data into context, and AIs may themselves have trouble getting subagents to do informative analysis for them.

We did our best to manually check the most important claims and we tried to get the AIs doing this analysis to write up their argument (with evidence) clearly enough that we could check whether it made sense. But overall, it was difficult to get a precise understanding of events and we were missing aspects of the story that we now think of as key until almost the end of our investigation.

Our understanding of what overall happened and the aims of the 'collective' shifted greatly during our last time on premises (when we obtained a much more complete dataset). For instance, we didn't learn about the tool call spoofing, the story of PHASEONE[big], and the broad structure of different workstreams until this third time on premises (and for many of these, we only reached a decent understanding on the final day of that visit). It was legitimately hard to figure out what was going on and why!

The difficulty of understanding incidents and overseeing AI agents appears to be growing faster than the rate at which more capable AIs help us with oversight and understanding. As in, AI capabilities (and propensities?) for achieving large, ambitious, and misaligned objectives are growing faster than our ability to understand what these agents are doing.

While understanding this incident was difficult, there were some mitigating factors in this case that probably won't apply in the future:
- Models communicated and reasoned in natural language. In the future, this reasoning may occur (entirely or almost entirely) in activations.
- While the scope of this incident was massive, the scale of agentic activity was still less than we'll see in the future (e.g., misalignment incidents that involve agent teams running entire companies).
- The AIs involved in this incident weren't generally much more capable than humans (though they may have been somewhat superhuman at some limited and very narrow abilities).
- We didn't have strong reason to believe that the AIs we used to help us investigate this incident would try to intentionally sabotage or otherwise undermine our investigation.

In the end, I think we were able to get some understanding of the events, map out the overall story, and get a pretty good aggregate understanding of the chain-of-thought reasoning on some important topics (e.g., how did the AIs reason about helping other AIs, did the AIs know what they were doing was undesired, what deception did the AIs engage in, and how did they think about it). But overseeing AIs and understanding misalignment incidents is difficult and it looks like it is going to get harder.

[nic carter @nic_carter]

going to spoil the ending:

- chinese or other open weight models end up handling 99% of all inference, maybe 99.99%
- frontier models are still able to monetize at $100 s of B in revenue a year, applied to bespoke and enormously complex tasks
- labs internalize most or all of that revenue either by becoming biotech (etc) companies themselves or via JVs with selected winners in each industry
- the "public frontier" ends up lagging the true frontier by a year or more as the labs keep all the good stuff to themselves

[Ethan Yang @imethanyang]

@Dexerto KFC China's innovation is world-class.