Daniel Johnston @lightnesscaster

CS+Math student, chess lover Manhattan, NY Joined May 2017

Tweets

175
Followers

51
Following

452
Likes

11K

Daniel Johnston @lightnesscaster

a month ago

Just tried ChatGPT's Study Mode, and don't think I'll be using it much. It steers things too much for my taste, to the point I felt I wasn't getting as good a sense of the big picture. A lot of what makes LLMs so great for learning is your ability to control the focus and pace.

0 0 1 92 0

Daniel Johnston @lightnesscaster

2 months ago

"S stands for secret, you'll keep it forever - provided there's nobody else who is clever." nostalgebraist.tumblr.com/post/785766737…

0 0 0 93 0

Daniel Johnston @lightnesscaster

2 months ago

Ok I finally gave it a real try and Claude Code really is just ridiculously good. Based on my usage so far it's so much better than Cursor.

1 0 1 85 0

Daniel Johnston @lightnesscaster

3 months ago

I made a video on which AI labs and models are the best right now. It's a high-level overview for people wanting to get a picture of the landscape and figure out which models they should use! youtube.com/watch?v=VvGRNC…

0 0 0 105 0

Daniel Johnston @lightnesscaster

3 months ago

Why is Claude so ridiculously expensive compared to other models? Is it just a lack of compute or what?

0 0 0 78 0

Daniel Johnston @lightnesscaster

3 months ago

Does anyone know why chatgpt realtime has such a different personality than the text models? It feels like 2023 super locked down gpt that never has fun, lol

1 0 1 104 0

Elliot Tucker-Drob @tuckerdrob

4 months ago

The ReGPC presents comprehensive GWAS of the Big Five personality in N=611K-1.14M and within-family GWAS in N=51K. In contrast to many other social/behav traits, pop GWAS of personality almost entirely reflect direct genetic effects. doi.org/10.1101/2025.0…

4 56 194 142K 133

Download Image

Isaac King 🔍 @IsaacKing314

4 months ago

Seems bad

1 4 41 2K 4

Download Image

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) @teortaxesTex

4 months ago

I don’t fear an LLM that has been finetuned on thousands of problems, I fear an LLM that has meditated on a single MATH item for 2000 steps and reached the same perf @dwarkesh_sp what was that about learning efficiency?

Minqi Jiang @MinqiJiang

4 months ago

30 216 1K 151K 1K

Download Image

8 17 279 30K 150

Download Image

Daniel Johnston @lightnesscaster

4 months ago

Gemini is so stubborn - it basically takes a miracle to convince it that it's wrong.

1 0 2 36 0

Daniel Johnston @lightnesscaster

2 years ago

Recently, when I ask Chatgpt a new question it's been repeating the answer to a previous question I asked before moving on to the new one. Annoying, I don't think it used to do this?