Joined December 2014
- Tweets 8,993
- Following 1,402
- Followers 1,653
- Likes 46,781
447 Photos and videos
Tim Hua 🇺🇦 retweeted
Jun 5
a year ago, i thought that slowing down frontier ai capabilities research was necessary yet impossible. i still think it's necessary (to buy time for safety research). “impossible” is looking shakier.
Jun 4
Our internal data shows Claude is accelerating AI development—a possible path to recursive self-improvement, or AI autonomously building a more capable successor.
It’s happening faster than we thought, and the implications deserve greater attention. anthropic.com/institute/recu…
5
5
106
8,840
Fwiw if this was actually just Bessent getting scared, I think it makes me a bit more optimistic of the future?
10
659
What is it like to be a coding agent?
Normally you can't tell! The UI hides the tool calls, subagent actions, and reasoning traces behind a tidy UI.
So I vibe coded a transcript viewer for Claude Code and Codex that makes it easy to see just exactly what your agent is doing.
6
2
26
680
You can download the viewer here: github.com/tim-hua-01/cc_tra….
Make sure to prevent transcript deletion in your Claude settings via
{
"cleanupPeriodDays": 99999,
"showThinkingSummaries": true
}
And add this:
model_reasoning_summary = "detailed"
To ~/.codex/config.toml:
1
4
73
I will say though, if you don’t know me personally, you should probably have a coding agent check the code for malware (or just give it the prompt request in the repo.)
You should not be executing arbitrary code you found on x dot com the everything app on your machine lmao.
1
2
60
I remember this tweet very well.
If you’re building AGI, you should expect the political economy around it to be ugly.
1
1
23
1,195
Like this is just not a serious attitude
x.com/firstadopter/status/20…
2
15
344
Tim Hua 🇺🇦 retweeted
Reminder that the last 12 months are plausibly the least crazy months of AI for the rest of your life!
It's only getting crazier from here ...
2
25
314
7,561
Tim Hua 🇺🇦 retweeted
Gemini 3.1 Pro and Gemini 3 Flash have most qualitative behaviors set by SFT, not RL, contrary to my expectations!
New GDM interp research: SFT is a big deal for safety relevant behaviors.
We recently investigated root causes for some of Gemini’s behaviors. We were surprised to find that many behaviors actually came from the initial supervised finetuning stage, not later stages like RL!
🧵
2
7
57
5,207
Putting export controls on Mythos could “wake up” other governments.
I think this makes it much more likely that Mythos’s weights will be stolen this year (is there a market on this?)
The government will be very mad at Anthropic if this happens. But it’s possible that they…
4
1
57
3,977
Or everyone is asleep at the wheel still so
x.com/matthewclifford/status…
2
222
Tim Hua 🇺🇦 retweeted
I have a weird feeling -- and please note, my weird feelings are not always reliable -- that this may be the beginning of things starting to get weird.
Jun 13
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-myt…
109
105
2,077
137,244
…steal an OpenAI model first?
Not sure how the dynamics of this will play out.
9
432
Tim Hua 🇺🇦 retweeted
Jun 12
This is pretty crazy. How long until 10 of the 15 FrontierMath: Open Problems are solved??
Jun 12
Claude Fable 5 scores very well on FrontierMath: Tiers 1–4 (v2), reaching 87% on Tiers 1–3 and 88% on Tier 4. This continues a streak of Anthropic models improving rapidly at math.
1
1
18
1,854