David Yang

[David Yang @davidyang05]

@svlevine so cool!

[David Yang @davidyang05]

@HengyuF W name

[David Yang @davidyang05]

@stevenlu0 @jonathanstray @serinachang5 congratulations Steven!

[David Yang @davidyang05]

@zhiyuan_zhou_ congrats sir

[Jensen Huang @JensenHuang]

@ChaseLochmiller @OpenAI GPT-6 Astra, trained on ~100K+ NVIDIA Grace Blackwell NVLink72. From ChatGPT to o1 to Astra in 4 years.

AGI has arrived. Congratulations @OpenAI team.

400K GPUs coming online next.

[Steven Luo @stevenlu0]

I had an amazing time working with @jonathanstray @davidzhaiyang Miu @serinachang5 at @berkeley_ai @CHAI_Berkeley over the past few months and I learned so much from everyone on the team!

Check out our full preprint ⬇️
Paper: https://t.co/cAQb0bsXDL
Data: https://t.co/od7MVRzixj

[Jan Zilinsky @janzilinsky]

I really enjoyed reading this paper.

Lots to like here (might write something longer later), but one fascinating part was how often people, regardless of ideology, said they approved of the responses from chatbots - even when reading the LLM text meant being exposed to arguments that “your own side might be wrong”!

Now, a small part of me wonders whether the high (often 70%+) approval rates may have partly been achieved because researchers told chatbots to be concise? It could be that users like texts about politics that 1) seem fair; 2) get to the point quickly.

Again, congrats to Serina, @jonathanstray, @davidzhaiyang @stevenlu0, and Miu Takagi on this great paper.

[Serina Chang @serinachang5]

When people strongly disagree on an issue, can they agree on what makes a good AI response?

We find: yes, more than you might expect!

We present PARETO, a large human study w >200k evals, measuring the Pareto frontier of approval btwn opposing groups on controversial issues 🧵

[David Yang @davidyang05]

@_chenglou ding dong oolong milk tea

[David Yang @davidyang05]

@chontang @ruima Ayaan Haque (UC Berkeley) is a researcher at OpenAI, and he was recently featured in the ChatGPT Image-2 Release!

[David Yang @davidyang05]

@chontang @ruima Not at AI2, but compensation at XAI should be competitive. IDK. Ask Elon?

[David Yang @davidyang05]

@chontang @ruima Actually, there are people in the MET undergrad program working at AI2 and XAI.

[David Yang @davidyang05]

@_avichawla Why would a candidate bring up flash attention as something to use in long context extension when it should be enabled by default during pre-training?

[David Yang @davidyang05]

@Diyi_Yang @percyliang lol open ai

[David Yang @davidyang05]

@MIT_CSAIL @a1zhang How do you see RLMs being adopted in a multi-agent setting?

[David Yang @davidyang05]

@zhijianliu Really cool!

[Jelani Nelson @minilek]

First Computer Science professor to run for California governor?

Maximum Flow, Minimum Waste

(image stolen from Reddit)

[David Yang @davidyang05]

@natolambert Awesome stuff

[David Yang @davidyang05]

@khoomeik @ExaAILabs

[David Yang @davidyang05]

Congratulations Steven! I've learned so much working by your side this past year. All the best in your future endeavors!

[Steven Luo @stevenlu0]

finally made it official while waiting in the airport in boarding group 6! I’ll be starting a PhD at @SCSatCMU in the fall, excited for the journey to come 🥳🎉