Mockapapella

807 posts

Mockapapella banner
Mockapapella

Mockapapella

@Mockapapella

10x Vibe Programmer 0.1x Vibe Debugger Please predict the next token responsibly

Katılım Eylül 2014
122 Takip Edilen50 Takipçiler
Mockapapella
Mockapapella@Mockapapella·
Wait. This wouldn’t happen to be a full GPT-2 XL pretrain, would it? I remember playing with one a couple months before OpenAI released the official weights. I think it was from some students in Texas and costed something like $40K in equivalent compute costs? Because if that was you, I owe you a huge thank you, because it was a part of the catalyst that helped me start my first startup, an automated essay writing service. Because of that (and an insane amount of grinding) we managed to get an LLM API up and running for our own service before OpenAI had even announced theirs (beat them by like a week or two I think).
English
1
0
0
112
Mockapapella
Mockapapella@Mockapapella·
This is almost becoming a daily occurrence at this point
Mockapapella tweet media
International Cyber Digest@IntCyberDigest

‼️🚨 MAJOR IMPACT: AI just found an 18-year-old NGINX critical remote code execution vulnerability. It has been disclosed on GitHub including PoC code. - Affects NGINX 0.6.27 through 1.30.0 - Triggered via the rewrite and set directives in config - Update NGINX ASAP - NGINX is a widely used HTTP web server, be sure to check its prevalence in other products

English
0
0
0
29
Mockapapella
Mockapapella@Mockapapella·
@rezoundous I do. Any time I downgrade I regret it due to the unreliability.
English
0
0
0
15
Tyler
Tyler@rezoundous·
Am I the only one using GPT-5.5 on xhigh all the time?
English
303
9
689
77.3K
Mockapapella
Mockapapella@Mockapapella·
At what point does AI saturate the field of Software Engineering and move onto the next field?
English
1
0
1
18
Mockapapella
Mockapapella@Mockapapella·
Fun little anecdote: a friend of mine works in manufacturing. Discovered Claude Code last fall. Went to a company engineering conference where the presenter heavily advocated for Copilot, and as such is the only officially approved agent harness they allow. Every single engineer that uses agents there still uses Claude Code. I've tried pushing him to use Codex, and to his credit he's given it a shot, but there's something about Claude that's just sticky. I think it's a combo of: 1. Claude has the best personality out of any model 2. Claude is the only model that is officially served from all 3 big clouds natively, so companies can be sure that their data stays within their cloud of choice I'm inclined to believe it's directionally correct
English
1
0
1
27
Mockapapella
Mockapapella@Mockapapella·
@sama If the model is slow I just run more of them in parallel
English
0
0
0
19
Sam Altman
Sam Altman@sama·
i get some anxiety not using the smartest-available model/settings. but sometimes i dont mind if it's really slow. i wonder if we should focus more on a price/speed tradeoff relative to a price/intelligence tradeoff.
English
2.1K
175
6.2K
609.3K
Mockapapella
Mockapapella@Mockapapella·
You vs the guy she tells you not to worry about
Mockapapella tweet mediaMockapapella tweet media
English
0
0
1
12
Sachin
Sachin@sachpatro97·
The Claude Code x Codex split terminal inside Cursor setup has me looking for an updated monkey with AK-47 meme
English
1
0
2
55
Mockapapella
Mockapapella@Mockapapella·
New experiment: Told codex "/goal make me $5 and do something you’re really good at" It's first inclination was to make a landing page offering a code review for $5 I do not have high hopes lol
English
0
0
0
40
Mockapapella
Mockapapella@Mockapapella·
@alex_whedon Very interesting, though I've heard similar claims before. Signed up for the preview, would love to try it out for some development work!
English
0
0
1
205
Alexander Whedon
Alexander Whedon@alex_whedon·
Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. @subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.
English
1.5K
2.9K
23.1K
12.6M
Mockapapella
Mockapapella@Mockapapella·
@koltregaskes Not surprised. 5.2 is still my go to for engineering. It thinks wide. No other GPT model thinks like that.
English
0
0
1
18
Kol Tregaskes
Kol Tregaskes@koltregaskes·
GPT-5.2 tops Factory's code review benchmark at 60.5% F1 on real open-source repos. The evaluation analysed 50 pull requests from repos including Sentry and Grafana against a human-curated golden set of bugs using consistent prompts and an LLM judge for F1 scores that balance precision and recall. GPT-5.2 even beat newer models such as GPT-5.5 at 47.9% F1 while budget models like MiniMax M2.7 delivered strong results at ten times lower cost and the full dataset is open-sourced on GitHub. factory.ai/news/code-revi…
Kol Tregaskes tweet media
English
5
1
15
867
Jonathan Grahl
Jonathan Grahl@jonathangrahl·
How do I stop Opus 4.7 from writing shit like this.
Jonathan Grahl tweet media
English
77
4
249
59.3K
Mockapapella
Mockapapella@Mockapapella·
Scooby-Doo, but Shaggy unlocks goblin-mode GPT-5.5 and hard-carries every mystery while the gang just stares.
English
0
0
1
76
Mockapapella
Mockapapella@Mockapapella·
@TheRealAdamG It does not think wide like GPT-5.2 does, and despite all of the version bumps, I find myself going back to it for engineering for the reliability it offers.
English
0
0
0
535
Adam.GPT
Adam.GPT@TheRealAdamG·
I've started to notice a repeating pattern with GPT-5.5 -- it's consistently pretty awesome at almost everything. Concerning.
English
59
33
1.1K
44.5K
maria
maria@maria_rcks·
A lot has gone wrong with Claude over the years. Here's every instance I could find Introducing clawd[.]rip
English
82
139
1.9K
272.7K
Mockapapella
Mockapapella@Mockapapella·
LLMs can read text, see images, and hear audio. But how would you give it a sense of smell?
English
0
0
1
13
Mockapapella
Mockapapella@Mockapapella·
For the second time in earnest, I have seen phenomenal writing come out of an LLM. Both from GPT-5.4 Pro. It seems to be relegated entirely to research reports after a detailed back-and-forth conversation.
English
0
0
1
25

Keşfet

@moinnadeem @rezoundous @sama @sachpatro97 @alex_whedon @subquadratic @koltregaskes @elonmusk