Sabitlenmiş Tweet
Randall Bennett
28.1K posts
Randall Bennett
@randallb
Ship more, build less. Confident AI builder @boltfoundry (Journo → coder → vidpresso (YCW14) → sold to FB) whatsapp: 16466701291
New York, NY Katılım Şubat 2007
418 Takip Edilen7.3K Takipçiler
Randall Bennett retweetledi
Doing some experiments today with Opus 4.6's 1M context window.
Trying to push coding sessions deep into what I would consider the 'dumb zone' of SOTA models: >100K tokens.
The drop-off in quality is really noticeable. Dumber decisions, worse code, worse instruction-following.
Don't treat 1M context window any differently.
It's still 100K of smart, and 900K of dumb.
English
Randall Bennett retweetledi
Agents don't want to sign up for hundreds of different accounts and maintain API keys and balances.
They want one unified skill that lets them access all the APIs with one balance.
Orthogonal. @orthogonal_sh
Excited to back these incredibly strong founders, @chrisspickett and @berasogut1.
English
Randall Bennett retweetledi
Guys, relax, the only risk in transiting the Strait of Hormuz is that Iran will shoot a missile at your ship. Otherwise it's open for transit.
An all-time quote from our Secretary of War.
English
Randall Bennett retweetledi
this is actually insane
> be tech guy in australia
> adopt cancer riddled rescue dog, months to live
> not_going_to_give_you_up.mp4
> pay $3,000 to sequence her tumor DNA
> feed it to ChatGPT and AlphaFold
> zero background in biology
> identify mutated proteins, match them to drug targets
> design a custom mRNA cancer vaccine from scratch
> genomics professor is “gobsmacked” that some puppy lover did this on his own
> need ethics approval to administer it
> red tape takes longer than designing the vaccine
> 3 months, finally approved
> drive 10 hours to get rosie her first injection
> tumor halves
> coat gets glossy again
> dog is alive and happy
> professor: “if we can do this for a dog, why aren’t we rolling this out to humans?”
one man with a chatbot, and $3,000 just outperformed the entire pharmaceutical discovery pipeline.
we are going to cure so many diseases.
I dont think people realize how good things are going to get
Séb Krier@sebkrier
This is wild. theaustralian.com.au/business/techn…
English
@dexhorthy @thdxr i’ve got a fix. build your repo like you build an efficient organization/company. keep automating yourself out of jobs instead of hiring yourself out of them.
this is basically the outsourcing to india problem but with slot machines instead of timezone fatigue.
English
The funniest part of this whole "CEOs who code" thing is that we're about 6-12 months from CEOs who just act like CEOs, and coders who act like CEOs.
Code is downstream of intent. Anything a CEO does that isn't communicating intent is a waste of time.
English
The AI safety people might be full of crap a lot of times, but I'm really grateful they exist because I think AI is in a much better position than it would have been without them.
Anthropic and OpenAI both existing (I think) is net good for the world.
English
Has anyone analyzed all of the technologies that came out of YC funded companies? Here's two off the top of my head:
Docker containers (dotcloud)
GPT (lol duh)
There's gotta be dozens of others that are hugely impactful.
English
@Austen having your own model is such a larp. there’s so much room at inference time to make a dent.
English
I’m sure Cursor _wants_ its own coding model, but to come from behind and beat OpenAI and Anthropic you’ve got to overcome a 5-10 year deficit against companies that have raised $50 Billion+ more than you and are swallowing everyone’s data.
I admire the ambition but tall order.
Yifan@YifanBTH
Seriously, Cursor needs their own model so badly
English
More than ever: Since you can create anything, create your own thing first so you can understand what's good or bad.
Unless, you have an urgent need and don't need to understand. Then use the easy thing.
But knowing how something works means you'll always have better taste.
English
ding ding ding!
I usually ask my assistant "without calling any tools, how would you do x" and then follow its path and then fix things that should be (but aren't) obvious.
dex@dexhorthy
yes I read the paper but if you are getting zero value from CLAUDE.md that’s a skill issue there are tons of very useful things you can use it for, they just might not be what you think Throw out the model generated file, hook up a proxy and inspect how it’s passed to the inference endpoint, and pay attention to what’s eating up context (trying to find the tests, looking in the wrong place) Write every token by hand here - you can do quite a lot
English
I will concede the singularity is here when this is no longer true.
Ben Garvey@bengarvey
Nothing makes you feel like "software isn't done yet" than paying a $4 convenience fee to pay your water bill on the shittiest website you've ever seen.
English
@swyx do you know if there's anyone in AI who doesn't think evals and reliability are the same thing?
English
This is why you can strip out 90% of people's system prompts and get better results generally.
They don't know how to deal with the edge cases so they bloat their prompt hoping for the best just to get the worst.
elvis@omarsar0
Don't overcomplicate your AI agents. As an example, here is a minimal and very capable agent for automated theorem proving. The prevailing approach to automated theorem proving involves complex, multi-component systems with heavy computational overhead. But does it need to be that complex? This research introduces a deliberately minimal agent architecture for formal theorem proving. It interfaces with Lean and demonstrates that a streamlined, pared-down approach can achieve competitive performance on proof generation benchmarks. It turns out that simplicity is a feature, not a limitation. By stripping away unnecessary complexity, the agent becomes more reproducible, efficient, and accessible. Sophisticated results don't require sophisticated infrastructure. Paper: arxiv.org/abs/2602.24273 Learn to build effective AI agents in our academy: academy.dair.ai
English
@auderdy This isn't the case with GLP-1s. Naltrexone is actually the anti-wanting drug. GLP-1s are ineffective for alcohol.
samhsa.gov/substance-use/…
thelancet.com/journals/eclin…
English
Real life Monkeys Paw situation — What do you mean you get your wish to become skinny/hot magically granted but loose your ability to fall in love??!!
Dr. Shin Geon-yeong (神建永), Ph.D.@asparagoid
We initially thought GLP-1s like Ozempic, Tirzapeptide and Retatutride just reduced food cravings. Now, we know they work for alcohol, cocaine, gambling and other addictions too But do you know what runs on exactly the same circuit? Falling in love GLP-1 receptors sit in the exact same brain regions that light up when you’re in love The insane thing about them is that they don’t just suppress appetite. They suppress wanting in general, including romantic craving another person Something like 60M+ people are now on anti-desire drugs and it happened in the blink of an eye I predict in the coming years, we will see people on these drugs be less able to fall in love. We will also see them fall out of love, or be unable to feel it, in relationships that were previously great If your girlfriend or boyfriend started taking GLP1s and your relationship started failing, there’s a good chance that’s why
English
If true, this means that model training should actually avoid internal knowledge and focus on problem solving.
IDK how you'd do that, but it seems like the right place to end up.
English
We should think of LLMs the same way we think of neuroscience:
Context window is short term / immediate consciousness.
Model training is instincts.
Long-term memory is tool calls.
English
@auderdy yeah... it's kind of a problem. Most of us are in recovery.
English
I asked my guy friend why men were obsessed with the a Roman Empire, and then 30 minutes later I’ve learned at the etymology of “Pyhrric victory” comes from from King Pyrrhus of Epirus, a Greek general whose army suffered irreplaceable casualties while defeating the Romans at the Battles of Heraclea like 280BC …and also that Romans once filled the Colosseum with water to enact sea battles, and there was also something called the Siege of Alesia in 52 BC where the Romans built a double ring wall to starve the inside people and protect against the relief army on the outside. Something else about filling a valley with logs so the army could cross. And also Sicily used to be a forest island but the Romans deforested it for all their ship building and battle supplies
English