moving from it's a prompt-skill issue to a "you didn't even ask" issue
David Gasca@gasca
The prompt for the Erdos problem was basically just stating the problem. yikes cdn.openai.com/pdf/74c24085-1…
English
David Gasca
9.3K posts
David Gasca
@gasca
🌲⛩️⛩️🌲 here for the AI Product @ Whatnot - ex @google, @twitter, etc.
California Katılım Nisan 2009
1.9K Takip Edilen8.1K Takipçiler
The prompt for the Erdos problem was basically just stating the problem. yikes
cdn.openai.com/pdf/74c24085-1…
English
David Gasca retweetledi
Today, we’re sharing that a general-purpose internal @openai model achieved a breakthrough on one of the best-known combinatorial geometry problems. Less than 1 year ago frontier AI models were at IMO gold-level performance. I expect this pace of progress to continue.
OpenAI@OpenAI
Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946. For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids. An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better. This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.
English
David Gasca retweetledi
One of my favorite Omni tests was turning my fern into a musical instrument with @AnezkaMin93 🎵🌱 Sound on 🔊
Google AI@GoogleAI
Reference anything: Gemini Omni extends Gemini's native multimodality, allowing you to blend combinations of text, audio, image, and video inputs into a high-quality, consistent video.
English
David Gasca retweetledi
Image editors, video resizers, custom shaders… if you can think of it, you can now build it. Introducing Tools in Google Flow.
Starting today, you can describe the tools you wish you had, and Google Flow will help build them for your workflows. Explore our Tools gallery for inspiration, including some from fellow creatives, share your own, or even remix existing Tools to fit your needs.
Learn how to get started in the thread below #GoogleIO
English
David Gasca retweetledi
Omni brings together an improved understanding of physics with Gemini's knowledge of history, biology, and culture, bridging the gap from photorealism to meaningful storytelling.
Actions have consequences, environments respond to events, and narratives evolve logically.
English
David Gasca retweetledi
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
English
David Gasca retweetledi
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically.
Instead focus on methods for creating knowledge that scale with computation, like search and learning.
English
Really enjoying the Spotify prompt-generated playlists... great feature
English
Was watching the Disney Imagineering Series and one thing I didn't realize is that Walt decided to build Disneyland after he tried to build a mini version in front of his studio in Burbank, CA and was denied by the Burbank City Council:
"A councilman stood up and looked at Walt and said, “We don't want a carny atmosphere in Burbank. We don't want people falling in the river, or merry-go-rounds squawking all day long.”
So instead he bought land in sleepy Anaheim...
English
yay!
Daydreaming a bit, but one thing I would like is to have agents that proactively go through and critique / offer suggestions on top of the design - kind of like "Design Crit" mode - e.g., here are 5 suggestions on the design that could make this better -> Accept/Deny/Modify vs. waiting for me to offer all the suggestions.
English
Cursor Design Mode + Multitask is magic
Cursor@cursor_ai
Use Design Mode in Cursor 3 to annotate and target UI elements in the browser.
Português
David Gasca retweetledi
Today we're sharing our work on interaction models. A new class of model trained from scratch to handle real-time interaction natively, instead of gluing it onto a turn-based one.
youtu.be/A12AVongNN4
YouTube
English
David Gasca retweetledi
Yes! "Models absolutely adore mediocrity!" and they're lazy!
I think people underestimate how much prompt engineering and model-wrangling is necessary to get best results
rohit@krishnanrohit
🚨 New experiment: I wanted to see how far AI could be useful as a researcher in a domain I don’t know well, where I couldn't just run tests and verify. As the Krishnan household is obsessed, I chose paleontology! Now, science in an unfamiliar field is hard because the output can be elegant, PhD-shaped, and completely wrong. Specifically I tried to figure out whether some of the theories I had about functional convergence were true. Like " when climate gets more volatile, do ecosystems become more similar?" ---- First, lessons learnt: 1/ models LOVE drifting into easier versions of your question over multiple rounds. You ask hard question A. It quietly solves tractable question A′. Then it returns a beautiful chart. 2/ You need to spend multiples of time verifying any response as asking questions. This needs huge amounts of process governance (since the answers cant be verified) and general sense checking (does that even sound right). 3/ You have to clean your workspace regularly. Models HATE deleting anything, and you need to shout at them ad infinitum to make this happen. Necessary to make later runs not confusing. 4/ Models absolutely adore mediocrity! They want to only do the inoffensive, unbtrusive, simple tasks that won't get them yelled at. Even when told not to! It's extraordinarily hard to get them to be bold. 5/ In a way it makes sense, because they have terrible attention to detail. They missed clades a lot or misstated the hypotheses halfway through. This is hard to fix when you don't know the domain, and requires constant vigilance! 6/ Models also think they can't do things that they can, and constantly say some analyses are a bad idea (see above re boldness). You have to tell them to stfu and calculate. They're so damn timid! 7/ Mainly for this but you HAVE to use multiple models to get the best results. They shock each other out of some basins and are useful tools to sense check. ------- The specific results were cool too btw. I did find out there is a "labour market convergence" during volatile times, where ecologies converged to different animals doing similar jobs at different times. And a predictino that you'd see more eg filter feeders during volatile times vs large chasing predators! Though this is true mostly in the Mesozoic! (I also disproved some theories like roles becoming "interchangeable" across clades.) --- I'd written a few years ago that "Analysts" are coming much in the way "Computers" became a machine. We're there now. Any curiosity you have can be analysed now by data. Any ideas you have about the world, you can now throw intelligence at it to test it out. It requires patience and practice, which is good, but I don't think we've nearly started to get to grips with this fact of reality! Essay: strangeloopcanon.com/p/the-spinosau…
English