albs — 3/staccs
@albfresco
Followers
1K
Following
6K
Media
2K
Statuses
8K
big tech big bro. applied AI.
YERRRRRR
Joined October 2010
major overhaul to pr arena dot ai. hope it looks more leaderboard-y
2
0
4
big E has earned the right to goon. he has like 12 kids and probably planted another 10 coming, and his companies will matter for the next 50+ years. he earned the right to goon. but you, not so much. so never goon.
0
0
4
for anyone using llms for code generation in custom setups (not via ide or platform), highly recommend using the diff / patch approach. it's fast, efficient, and effective. yes, you have to build some more overhead. but openai basically gives you the utils here.
e.g. learn how patches are actually generated, parsed and applied -
1
0
1
hi @ChatGPTapp, the shift + up/down arrow to jump around messages conflicts with when you're editing a past sent message, highlighting multiple lines. it makes edits slower. that experience is probably more common / important than jumping around messages.
0
0
0
is AI gonna be like search (just partner and license someone like google)? . or maps (wait until tech matures then build a completely custom thing years later than the rest)? . i think search.
Here’s what I would do if I was CEO of Apple:. -quadruple the RAM in iPhones to 32GB and have Max model at 64GB. Memory is oxygen for local on device AI. More = smarter and more powerful. Take the margin hit. Memory isn’t even that expensive. -buy Mistral or Anthropic and invest.
1
0
1
RT @EyalToledano: we've shipped taskmaster v0.23 🚀 . → official @taskmasterai @code extension .→ manage your taskmaster workflow visually.→….
0
42
0
youtube shorts™ coming to the Gemini app.
The worst part about vibe coding is sitting around waiting for the model to be done.
0
0
1
RT @albfresco: @paulg i'm not sure i will ever understand this. every day, i am shocked to understand how much of University "stuff" is fun….
0
3
0
memes.predict.the future.
0
0
0
wow. it's almost like openai knew what they were talking about. all good, big bros.
1
0
4
RT @albfresco: found the cure for Pedro Pascal’s social anxiety! I think it will be super effective!
0
1
0
i continue to be absolutely baffled by 4.1 performance. quiet, not verbose, no bullshit, just pure diffs getting applied and one shotting everything i ask. what the hell.
ok 4.1 has grown on me. particularly in vs code with copilot agent mode. it’s just very direct. it “thinks” when you ask it to. but if you just want the job done, it just does the thing with its best guess. and it’s generally pretty decent at it.
1
0
3
everyone knows subscription models don't work for inference. however, we must still offer them for now. it's a gentle, predictable, safe stepping stone for new users to try something new. it will eventually turn into a utility company model, but it has to keep happening for now.
Imagine if gas stations didn't tell you how many gallons you were getting. The station built your car, but you can't look under the hood. That's the inference subscription dilemma in one image, and the reality of selling commodities on subscription models. 🧵.
0
0
2
found the cure for Pedro Pascal’s social anxiety! I think it will be super effective!
1
1
2
claude code will be rebranded to be a generic agent framework. so will codex cli. there is no reason these systems need to be code specific. they do all the things any kind of agent would do. that's why these usage updates are needed. the time is soon.
0
0
2