Joined March 2017
- Tweets 3,499
- Following 97
- Followers 471
- Likes 5,635
32 Photos and videos
We recently fought two unnecessary decades-long wars of occupation. The invasion of Iraq was sold to Americans with falsehoods and exaggerations. Thousands of American and our allies' servicemembers died for no strategic gain in Afghanistan. Many of our current problems are a result of not sufficiently valuing the lives of American servicemembers, particularly the enlisted infantrymen who bear the heaviest burden of our wars.
Jun 12
The USA has joined the ranks of the world's post-heroic countries that cannot tolerate even a few casualties to defend even major interests. 82nd troops & Marines in place w ample air support to cut risks did not occupy the Gulf islands to keep the seaways open. WH said no
46
155
1,282
169,735
Tactics/os retweeted
Jun 13
What else could one expect from a US admin that canceled funding research on transgenic mouses thinking they were transgender?
Expose the same to Anthropic fearmongering and you get that.
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…
2
4
30
2,247
Jun 13
It's very funny how it's immediately clear to everyone who this is about...
Jun 13
33
Tactics/os retweeted
Jun 12
yeah sure must be this, and also maybe 3 years of open research and development
DeepSeek LLM (2023) arxiv.org/abs/2401.02954
DeepSeek-Coder (2023) arxiv.org/abs/2401.14196
DeepSeekMoE (2024) arxiv.org/abs/2401.06066
DeepSeekMath (2024) arxiv.org/abs/2402.03300
DeepSeek-VL (2024) arxiv.org/abs/2403.05525
DeepSeek-V2 (2024) arxiv.org/abs/2405.04434
DeepSeek-Prover (2024) arxiv.org/abs/2405.14333
DeepSeek-Coder-V2 (2024) arxiv.org/abs/2406.11931
DeepSeek-Prover-V1.5 (2024) arxiv.org/abs/2408.08152
DeepSeek-VL2 (2024) arxiv.org/abs/2412.10302
DeepSeek-V3 (2024) arxiv.org/abs/2412.19437
DeepSeek-R1 (2024) arxiv.org/abs/2501.12948
DeepSeek-Prover-V2 (2025) arxiv.org/abs/2504.21801
DeepSeek-V3.2-Exp (2025) github.com/deepseek-ai/DeepS…
DeepSeek-OCR (2025) arxiv.org/abs/2510.18234
DeepSeekMath-V2 (2025) arxiv.org/abs/2511.22570
DeepSeek-V3.2 (2025) arxiv.org/abs/2512.02556
DeepSeek-V4 (2026) huggingface.co/deepseek-ai/D…
Jun 11
Looking back, I never understood the hype of the "deepseek moment" from last year. Distilling others' models is possible and easier than pushing the frontier. Like, nobody remembers Alpaca?
16
50
647
36,716
Tactics/os retweeted
Jun 12
Must read by all those who think that a real neuron is equivalent to a single weight in a deep learning model.
Jun 11
What can a neuron compute?
Real biological neurons are complex, but how capable are they?
Using a new method, we found that a single cortical neuron can classify cats vs dogs, recognize spoken words, and solve 10-bit parity, all tasks thought to require entire networks. (1/15)
2
13
93
10,976
Tactics/os retweeted
Jun 11
kinda crazy that someone's full-time job was to steer claude to sabotage ML research capabilities for paying customers
71
161
3,494
140,659
Tactics/os retweeted
Jun 11
Asked a question so dangerous they sent my ass to haiku
78
394
8,981
209,911
Tactics/os retweeted
Jun 11
200k papers, but only one can reign supreme!
if you want to learn about how to juice results with bad baselines, all you need to do is sample a few optimization papers, especially those from the establishment of prestigious affiliations and the compute-rich (usually more cutthroat, more juiced results)
sometimes learning what _not_ to do can really help build taste
Jun 10
ChatGPT tells me there are ≈200,000 ML optimization papers written -- chatgpt.com/share/6a29abcf-f…
1
5
93
16,499
Tactics/os retweeted
Jun 11
i have too much disrespect for optimization research to comment on that
2
2
15
1,262
Tactics/os retweeted
Jun 10
I plan to live Anthropically. If someone asks me about something I don't like I'll just become a stupider version of myself
59
362
4,553
149,638
Tactics/os retweeted
Jun 11
spent the day with fable on a bunch of random stuff and it’s very spiky, imo.
it is brilliant in the same message where it makes 1-3 severe mistakes, which means you have to check even more stuff more in-depth, wasting time and tokens
10
2
92
5,027
NEW: Anthropic is walking back Claude Fable 5's policy to covertly degrade performance for competing AI researchers, after facing fierce backlash.
“We’re changing Fable 5’s safeguards for frontier LLM development to make them visible,” Anthropic tells WIRED. “We made the wrong tradeoff and we apologize for not getting the balance right.”
167
251
2,548
731,030
Tactics/os retweeted
Jun 11
things that are probably unwise:
- spending nonzero time on X arguing about credit attribution from a junior researcher's 2024 blogpost
- while holding onto trauma from being pulled out of the running for an L10 promo & even greater glory by prior-prior-employer
- while attempting to also start yet-another-neolab
priorities are questionable, judgement seems impaired, insecurities on full display (maybe that's humanizing?)
but altogether not a good sign from a "leader" anyone sane would want to collaborate with
13
9
373
35,310
Tactics/os retweeted
Jun 10
so much aura loss just for scraps
May 17
"optimizer research" is the most intellectual, big-brained way of avoiding grunt work
1
3
102
19,823
Tactics/os retweeted
Archimedes, the Token Maxxer.. #AIAphorisms
I am confused as to why it is surprising that throwing oodles of test time compute can increase the solvability horizon (modulo a good verifier in the loop).. wasn't this known even from as far back as depth-limited minimax?
1
14
1,573
Tactics/os retweeted
When a new model comes out, I like to give it its own system card and ask questions about it.
This does not work for Fable 5, as it routes me to Opus 4.8 for "safety" reasons...
11
14
313
49,549
Tactics/os retweeted
Jun 9
imagine telling your customers there's a small chance you'll randomly decide they're using your product wrong and you won't tell them but will secretly silently sabotage their work
41
206
2,988
107,975