Doc Xardoc @andrewb10687674

Joined May 2022

Tweets

3K
Followers

251
Following

6
Likes

111K

Doc Xardoc @andrewb10687674

20 hours ago

today in 'WHY CHATGPT????"

0 0 1 19 0

Download Image

Apparently ~800 lines of code is the point where GPT-5 starts making egregious errors (randomly removing functions, duplicating variables, weird formatting). This is disappointingly not noticeably longer than Gemini-2.5-pro

0 0 1 164 0

Doc Xardoc @andrewb10687674

a week ago

Does anyone speak sweedish? I asked nano-banana for ikea instructions showing how to survive a robot uprising

0 0 0 66 0

Download Image

Doc Xardoc @andrewb10687674

2 weeks ago

I'm very interested by the phrase 'in pixel' here, which implies either an open-source model or a closed source model that's going to have its weights leaked very soon.

Peyman Milanfar @docmilanfar

2 weeks ago

I'm very interested by the phrase 'in pixel' here, which implies either an open-source model or a closed source model that's going to have its weights leaked very soon.

46 68 1K 148K 140

0 0 0 64 0

Doc Xardoc @andrewb10687674

3 weeks ago

So excited!

Qwen @Alibaba_Qwen

3 weeks ago

So excited!

50 55 977 100K 84

Download Image

0 0 0 42 0

Doc Xardoc @andrewb10687674

3 weeks ago

I think it's been long enough that I can say... GPT-5 *thinking* is a meaningful improvement over o3. Every other version of GPT-5 (non thinking, speaking, mini...) is mildly retarded.

0 0 0 48 0

Doc Xardoc @andrewb10687674

4 weeks ago

oh to have the confidence of gemini 2.5 pro declaring the 'final definitive fix for the file' when on it's 100th iteration of me saying "Nope, you still didn't fix the bug"