Shailesh @0xThoughtVector

Product Manager. Just trying to survive each day as it is. github.com/gut-puncture Bengaluru Joined January 2010

Tweets

4K
Followers

266
Following

2K
Likes

60K

Shailesh @0xThoughtVector

4 hours ago

I'm glad more people feel the same. It's maddening to read fragments of sentences especially when you're trying to learn something you didn't know. I have explicitly stated "No phrases as sentences" in the system prompt, which does the trick.

Chris @chatgpt21

17 hours ago

1 0 12 1K 2

0 0 1 39 0

Shailesh @0xThoughtVector

13 hours ago

Finished '1984' and it is absolutely horrific. A perfectly designed oligarchy, leaving no avenue for free thought, and stripping its subjects from all of their humanity. Need to think much more about this. Will take a break from my plan and read 'Sapiens' next. Long overdue.

Shailesh @0xThoughtVector

a week ago

0 0 0 52 0

0 0 0 28 1

Shailesh @0xThoughtVector

2 days ago

It's crazy how lazy people are with prompting models. It's one line ambiguous slop most of the time with no preferences, details or caveats. Losing out on so much intelligence.

0 0 0 20 0

Shailesh @0xThoughtVector

2 days ago

This is why you don't compare heavy-tailed domains with low variance domains

eloise @dundereloise

3 days ago

This is why you don't compare heavy-tailed domains with low variance domains

25 15 869 46K 136

Download Image

0 0 0 26 0

Shailesh @0xThoughtVector

2 days ago

If an Open Weights model is not really open source and you can't really do anything with it, why not release Sonnet 3.5 to the public. We'll care for our dear old friend. Fucking Anthropic.

林徐乐 Lin Xule @LinXule

3 days ago

If an Open Weights model is not really open source and you can't really do anything with it, why not release Sonnet 3.5 to the public. We'll care for our dear old friend. Fucking Anthropic.

0 2 15 34K 1

Download Image

0 0 0 42 0

Shailesh @0xThoughtVector

2 days ago

Very soon we're going to see large AI start-ups taking Ls. Cursor will be first, Perplexity next and then mighty Anthropic.

0 0 0 22 0

Shailesh @0xThoughtVector

2 days ago

When you use the Claude.ai website, it feels so smooth and fast. You just know no one's using the website.

0 0 0 13 1

Shailesh @0xThoughtVector

3 days ago

The Open AI definition of AGI is bull-shit. We've already achieved AGI. It's jagged just like human beings.

1 0 1 32 0

Shailesh @0xThoughtVector

3 days ago

Would be very interested to know Cursor's numbers now. It seems everyone has moved away from it for CLI tools. And I don't think they were ever a normie product like VS Code + CoPilot (MS distribution) and Lovable (non-coder normie sub for coding tools).

0 0 0 27 0

Shailesh @0xThoughtVector

5 days ago

Orwell argued that the Proletariat would recognise oppression if they were educated. But most educated people are pacified by the inability of their own minds to think critically.

0 0 0 23 0

Haider. @slow_developer

7 days ago

openAI simplified the model lineup instead of o4, o4 mini, o3, o3 mini, and 4o, you now get: GPT-5 GPT-5 auto GPT-5 thinking (standard, extended) GPT-5 pro (light, standard, extended, heavy) GPT-5-mini GPT-5-nano GPT-5 codex model name Apocalypse

73 53 1K 57K 79

Shailesh @0xThoughtVector

6 days ago

To and fro technical conversations with models are effectively dead with GPT-5. Best I can do is talk with Auto to finalise a prompt which can be used to run Extended Thinking. It's just so slow.

0 0 0 39 0

Shailesh @0xThoughtVector

a week ago

This is... not a joke?

Area @areatechnology_

a week ago

This is... not a joke?

195 101 2K 1.2M 1K

Download Image

0 0 0 84 0

Shailesh @0xThoughtVector

a week ago

Often when I want a detailed answer from LLMs and they keep being lazy, it's an issue of a low output token limit. I should be able to set unlimited output lengths, at least in API. This is an issue with pricing all tokens alike. I would happily pay more for longer outputs, tho

0 0 0 23 0

Shailesh @0xThoughtVector

a week ago

stupid question, but if Open AI could reduce o3's cost by 75% bcz of an optimisation, why can't they do it for all their models?

0 0 0 23 0

Denny Zhou @denny_zhou

a week ago

we live in a parallel universe lol

79 82 2K 359K 1K

Shailesh @0xThoughtVector

a week ago

Current codex makes me feel, more than anything previously, that most coding work is solved. It just works, it seems!

0 0 0 29 0

adi @adonis_singh

a week ago

dont worry guys gpt-6 will unite all of this!

SIGKITTEN @SIGKITTEN

a week ago

dont worry guys gpt-6 will unite all of this!

29 18 852 103K 63

Download Image

16 17 819 52K 41

Shailesh @0xThoughtVector

a week ago

"Talking to her, he realized how easy it was to present an appearance of orthodoxy while having no grasp whatever of what orthodoxy meant." Seems to apply to a lot of Indian religious youth too. Starting reading '1984' btw. Intense. Horrific. Thought-provoking.

Shailesh @0xThoughtVector

3 weeks ago

0 1 1 96 1

0 0 0 52 0

Shailesh @0xThoughtVector

2 weeks ago

Is it weird that the tools to pacify the proletariat (TikTok, Instagram, TikTok) became controlled by capitalist entities rather than governments, or is his expected in late stage capitalism?