Leary

And Sonnet 5.5 is more expensive than Opus 5.5 to hit that score on terminal bench!

Uploaded my code base and it forced switched to Opus 4.8 after thinking for 5 minutes even though I prompted it to not work on cybersecurity related things. Amazing.

news from 2025

“Large frontier developer”: a frontier developer that together with its affiliates collectively had annual gross revenues in excess of five hundred million dollars ($500,000,000) in the preceding calendar year.

So any large company training LLMs, no matter the capability, is considered a frontier developer?

Except that does make sense because the new country would be the weighted average of the average CO2 contribution for both Americans and Nigerians.

Amazing analysis, except for the fact China was dirt poor before communism.

No way. I read Noah Smith (number 1 economist on substack) every week and he says one must measure China in per capita terms for all the good stuff: GDP (preferably nominal), and in aggregate terms for all the bad stuff (pollution, carbon dioxide emissions).

Also, it's logically impossible for China to be good. I have found a mathematical proof:

1: Democracy is good

2: China is not a democracy

Therefore, obviously China is not good.

But how are you gonna foresee the total annihilation of the Chinese semi industry without Jordan the necromancer interpreting party documents for you?

Does anyone know which technology on this tree has the most descendents?

METR of only 2 hours and 15 minutes. Fast takeoff less likely.

GPQA Diamond: gpt-oss-120b: 80.1%, Qwen3-235B-A22B-Thinking-2507: 81.1%

Humanity’s Last Exam: gpt-oss-120b (tools): 19.0%, gpt-oss-120b (no tools): 14.9%, Qwen3-235B-A22B-Thinking-2507: 18.2%

Did it with https://sites.google.com/site/jiaxiongyao16/nighttime-lights...

USA (2013-2023 CAGR: 2.3%) 2014: 6.2% 2015: -5.3% 2016: -1.8% 2017: 15.2% 2018: -4.9% 2019: 4.5% 2020: -5.4% 2021: 6.7% 2022: 14.5% 2023: -3.6%

China (2013-2023 CAGR 7.9%) 2014: -1.7% 2015: -1.2% 2016: -5.1% 2017: 53.3% 2018: -1.0% 2019: 7.5% 2020: 6.5% 2021: 11.4% 2022: 4.2% 2023: 10.8%

0:10 into the video, that's light skinned to you?!

Translation:

You provide the capital and the data, we'll co-own the data centers share the models until Trump and the US government decide to shut it off as a bargaining chip.

They are not "English-first". Deepseek-R1, for example, reasons in Chinese when you ask it a question in Chinese.

Nah, ByteDance will sell if the US lets Nvidia sell them some B200s

So OpenAI's ChatGPT output cannot be copyrighted and it's legal to distill it?

I don't think they are claiming 40% of males today have y-haplogroups descended from the Yamnaya

r1b's population is only 190 million [1]

[1]https://www.razibkhan.com/p/the-haplogroup-is-dead-long-live...

Maybe because real estate prices are not captured by CPI, only rental prices. People are taking longer and longer to save up for a house[1].

https://www.crews.bank/blog/real-estate-prices-vs.-income

The authors make a case that China's GDP is overestimated from past night time studies so I ran an analysis on which country produces the most nighttime lights.

110 RUS 926043522.5

136 USA 816841431.0

24 CHN 541388528.0

21 CAN 537446163.4

16 BRA 259637239.0

62 IND 229451980.0

4 AUS 173842078.0

63 IRN 152673182.8

84 MEX 116989787.0

3 ARG 110292451.1

Russia is the real secret superpower?

On an EV basis, the worst thing SBF did was probably funding Anthropic and accelerating the demise of humanity.

One of the people on the Qwen team tweeted this instruction.

To test: https://chat.qwen.ai/ and select Qwen2.5-plus, then toggle QWQ.

How does this compare with Grok 3's parameter count? I know Grok 3 was trained on a larger cluster (100k-200k) but GPT 4.5 used distributed training.

As recent as 2017.

"Earlier this year, I noticed something in China that really surprised me. I realized I felt more comfortable discussing controversial ideas in Beijing than in San Francisco. I didn’t feel completely comfortable—this was China, after all—just more comfortable than at home."

https://blog.samaltman.com/e-pur-si-muove

He also mentions giving a talk for YC in China in 2018:

https://blog.samaltman.com/idea-generation

This is weak sauce. America should limit chips to other countries and then make Taiwan sell them to the US with the 100% tariffs.

Oh wait, that's what the AI diffusion rules do.

Enjoy this poem:

I dwell in Possibility – (466)

By Emily Dickinson

I dwell in Possibility –

A fairer House than Prose –

More numerous of Windows –

Superior – for Doors –

Of Chambers as the Cedars –

Impregnable of eye –

And for an everlasting Roof

The Gambrels of the Sky –

Of Visitors – the fairest –

For Occupation – This –

The spreading wide my narrow Hands

To gather Paradise –

Dario certainly had no qualms about working with authoritarian regimes when he worked for Baidu!

Also, if the US seeks a permanent advantage over China by 2027 with self-replicating AI, so much so that it can dictate whatever terms militarily. Wouldn't that just force China to send a few cruise missiles to TSMC before that happen?

Does this mean when you use OpenAI as an enterprise customer, they can see exactly the queries and answers? So much for privacy!