VC rounds now asking how many companies your chatbot has hacked and declining if it's less than 3

Exciting to announce Discovery Loop II. While many other companies simply focus on automating research, our company focuses on automating the launching of automated research companies. By next year we hope to be launching one of these every minute of the day.

If it's any comfort, I also find the rate of change of everything somewhat frightening. Exciting, but also scary. The future will clearly look very different and it's hard to imagine how.

I actually configure AWS correctly now due to embedded chatbots

Maria Antoniak@mariaa.bsky.social · 11h ago

Is AI useful... Well I've avoided Figma for years because every time I opened it, it looked too annoying to learn. But now they have a chatbot, and I made three websites yesterday.

This website is desperate to argue with someone who disagrees with them and in its absence, will find the nearest neighbor to that

To whoever wrote this, thank you for saying it and know that it helps a bit. And I'm still so sorry and frustrated that we have silenced international students.

If your vision of the future doesn’t involve productivity growth, it’s a vision of many people living substandard lives

Systems that do not have regulatory barriers or gatekeepers to AI are rapidly getting overwhelmed / transformed: mathematics, paper writing, code, etc. For everything else, it'll be slower but it makes sense to start planning now!

as we get further into the end of days, I'm reminded about how much of research taste, values and judgment is bound up in the taste, values and judgment of small communities of researchers

Not here because it's a left-leaning site, here because I see this as a renewed vision of the internet.

My radical suggestion for peer review: Move from nominal pre-publication review to explicit post-publication review. The original reason for peer review was that journal pages are a limited resource, so we need a filter before publication. That no longer makes sense with digital publishing. 1/

🚨🚨🚨🚨🚨

robert j bennett, ceo, results omega@robertjbennett.bsky.social · 2d ago

hey! that’s today!!

Left, right, whatever, I just want graduate students posting their papers here

Magic Metal Montana, by Charli XCX, is one of the better Strokes songs in recent years

LLMs adopts linguistic quirk -> Humans avoid linguistic quirk -> Humans develop new linguistic quirks -> LLMs adopt linguistic quirk... LLMs can update a lot faster though, and so I wonder if this all does something quite weird to written english.

Are policy gradient methods hopeless without other machinery? Deep learning works well when the Hessian is well conditioned. But the policy objective is under no obligation to give us that. And no amount of architecture engineering can rescue us from problems w/ the objective.

I’m saying this as a routine voluntary user of LLMs: constant questioning and re-evaluation of whether you are interacting with a human does real psychic damage.

Golden Gate Claude was actually the peak of human civilization and it has all been downhill since then simonwillison.net/2024/May/24/...

An algorithm that makes the people who interact with it better (according to their own stated values) rather than worse is worth pursuing. And, if it turns out to be impossible then it’d be a good idea to shut the whole thing down

A lot of mechanisms for getting reviews are increasingly punitive. But reviewing a paper you care about is a joy and a good research exercise. The core of the problem is having to review papers you don't care about.

If you’re curious about why people are so excited/concerned about the state of agents, this is a really nice tutorial. If you think this stuff is useless and you’re open to changing your mind, this might be interesting!

David P. Reichert@david-p-reichert.bsky.social · 3d ago

Many don't see how much AI has changed recently. Coding agents aren't just chatbots. We can’t deal with the challenges of AI if we don’t understand it. Here's a post to provide some evidence (Claude doing small ML experiments) and form intuitions. davidpreichert.substack.com/p/if-you-hav... 🧵..

Another fantastic paper from Valeo folks, 35M mid-simulator kilometers to develop safe behavior under partial observability valeoai.github.io/Pictura/

Pictura: Perspective-View Self-Play at Scale for Driving

Self-play driving policies trained directly from rendered perspective images, without privileged vectorized observation of the surroundings.

valeoai.github.io

It's interesting that so much dev later and LLMs still try to try and except their way out of failures by default while "don't have silent errors" is among the top rules of good programming

Be kind to the academics in your life and try to help them pretend it's still July

"We're all worried," as what it means to do research (in my field, Theoretical CS) seems to be shifting, and shifting fast. What to do? Senior researchers must lead by example, knowing that not everything will pan out. What I'm suggesting below may not work everywhere, but here's my own advice: 1/

Is there a good LLM-based language immersion app yet? Gives you feedback on mistakes, lets you run through any topic, answers questions, etc?

Did you know there’s actually an oversupply of brilliant academics so you don’t have to tolerate the monsters

Eugene Vinitsky 🍒@eugenevinitsky.bsky.social · 3d ago

Talked with a friend about her PhD experience and oh my god, some fields need clearing house. Just burn the whole thing down and start over

Talked with a friend about her PhD experience and oh my god, some fields need clearing house. Just burn the whole thing down and start over

Another overview of one of the astra solutions worth checking out

Chris Peikert@chrispeikert.bsky.social · 3d ago

1/ Initial reactions after some hours with this groundbreaking result proving the NP-hardness of poly-approx CVP/NCP: It is most likely correct, but more importantly, it is original, elegant, and beautiful! (Also: it is easy to improve, quantitatively.) openai.com/index/ten-ad...

This week's #PaperILike is "Algorithm Runtime Prediction" (Hutter et al., 2014). "I don't know the answer, but I know I will know the answer in about 10 minutes" (e.g., to a SAT question). Pretty wild that this is possible! PDF: arxiv.org/abs/1211.0906

Algorithm Runtime Prediction: Methods & Evaluation

Perhaps surprisingly, it is possible to predict how long an algorithm will take to run on a previously unseen input, using machine learning techniques to build a model of the algorithm's runtime as a ...

arxiv.org