Wais P (@waisp) profile picture

Definite Optimist, In love with machine learning.

San Francisco, CA
Joined January 2011
Photo by Wais P (@waisp)Photo by Wais P (@waisp)Photo by Wais P (@waisp)Photo by Wais P (@waisp)Photo by Wais P (@waisp)Photo by Wais P (@waisp)Photo by Wais P (@waisp)Photo by Wais P (@waisp)Photo by Wais P (@waisp)Photo by Wais P (@waisp)
Ripple quantizing itself -> a 30.3GB version that's pretty good!
One of the most important things in development of local coding agents is that we should be able to serve the modal locally -> there are many ways to get here but one of the most important techniques in serving is quantization. At the @RicursiveAI hackathon, I wanted to understand that whether we can let an agent loose to quantize itself i.e. can ripple 65B quantize itself. As fundamentally quantization is a search problem, you search across which layers to quantize and which layers to leave out. Thread 1/n 🧵
18
3
41
4,761
Wais P retweeted
i learnt this lesson this year. it's helped me follow through on my ideas and not get stuck overthinking it don't ask if you should do the thing. do it, then come back with a demo or something tangible you can just do things
210
744
101
9,805
371,917
Wais P retweeted
holy f*ck... SpaceXAI engineer Lauren Tan released her method for coordinating agents through two feedback loops: start with coordinate.prompt to prioritize work by cause, dependencies and impact → Bind each task to the code and evidence checked record the main SHA, source reports, reproduction steps and actual output keep suspected causes separate from confirmed observations. Before dispatch: → assign one owner and explicit edit paths → identify prerequisite tasks and shared interface changes → define the setup, command or user scenario used for acceptance → specify the output, state change or assertion that must pass group reports when evidence supports a shared cause retain their source links in the same brief if the checked revision or dependencies change, reassess the reproduction and acceptance checks before continuing verify.prompt (test the candidate) run the actual application and replay the affected workflow UI: complete the journey and inspect the result performance: compare equivalent baseline and candidate profiles quality: run types, lint and relevant tests Return five things: tested revision / reproduction / commands + exit codes / artifacts / unrun checks + blockers a failed check sends the change back for correction and another run a check that cannot execute stays NOT_RUN, with the missing prerequisite and next action recorded → Use garden.prompt on recurring corrections review work history and group findings by shared cause improve the structure or add an automated check. move stable mechanical steps into scripts, CLIs or codemods trial the change on real work and retain evidence of the failure it prevents or detects keep the original case so later changes exercise the same requirement expand delegation after repeatable checks. keep permissions, review rules, stop conditions and rollback options explicit Save this to spend less time coordinating fixes ↓
Opus 5.5 can become your own AI development team One GitHub issue becomes coordinated assignments, parallel work in Git worktrees and a draft PR with test results attached. Here’s how to build the team in 7 steps:
5
17
239
21,081
Wais P retweeted
This is f*cking insane. Anthropic engineer: "Prompting is basically over 95% of our engineers already run self-improving loops. The next move is graphs" In barely 10 minutes, she builds her entire Claude Code setup live, straight from an empty terminal Empty Terminal → Agents → Loops → Graphs → Self-Improving System Prompting was the old way. Engineering the system around the model is the new one Save it and watch it today Then read the article below ↓
49
137
9
1,712
358,956
At @SpaceXSI, Lauren Tan (@poteto) and Roman Ugarte (@romanugarte_) are on the core team that took Grok Bot from first line of code to launch in seven weeks. It lets you spin up AI teammates with their own identities and logins, and they just keep working in the background. Their marketing team gave a bot its own ad budget. Lauren built an entire feature with her bot in one Slack thread without ever opening Cursor. They joined us on the @LightconePod to talk about what it’s actually like working next to bots every day, and how the hard part is shifting from doing the work to figuring out what to do and why. 03:36 — Giving Bots Real Jobs 06:49 — One Bot or a Team of Specialists 10:39 — Sharing Bots Across the Company 14:46 — Identity and Permissions for AI Coworkers 18:37 — Bot-to-Bot Collaboration and Trust 22:35 — The Workflow Behind 4,000 PRs a Month 26:00 — Teaching Agents to Write Code You Trust 29:01 — Building a Personal Software Factory 33:44 — The Future of AI Coworkers 36:47 — Rethinking the AI-Native Company 40:13 — Teaching Agents to Check Their Own Work 43:56 — How Much Code Review Is Enough? 47:33 — Codebase Guardrails That Save Tokens 52:20 — When Product Judgment Becomes the Bottleneck 56:29 — From Finance to AI Engineering
132
237
56
2,375
700,269
"The person who writes down the thing has tremendous power. One of the ways you find the up-and-comers at a tech company is by seeing who wrote down the plan. That doesn't mean they came up with everything and it doesn't mean they had all the ideas, but it means they're actually able to organize their thoughts — and have the energy, motivation, and skill to communicate it in a written form."
4
15
6
131
21,951
Last day of the very special @COLM_conf !! Surprise, surprise, I am actually here to present a poster, than just tweet 😆 Stop by poster #3 this afternoon if you want to learn about training LMs entirely, and from scratch, on knowledge graphs! Why, How and What we learned.
7
86
4
605
97,802
if you know what you’re doing, Claude + Cursor combination is killer what used to take days to build has been reduced to hours if not minutes for me.
27
66
6
1,144
182,617
Mastering Time Management ‼️
65
12,314
266
86,090
12,312,143
All the startup accelerators you can apply for: 1. PearX ($250k-$2m) 2. Accel Atoms ($500k) 3. Antler ($200k for 8%) 4. Soma Capital ($100k) 5. Sequoia Arc (Variable) 6. a16z Speedrun ($750k) 7. LAUNCH ($125k for 7%) 8. OpenAI Converge ($1m) 9. Techstars ($20k for 6%) 10. Alchemist ($25k for 5%) 11. NEO ($600k Uncapped) 12. AngelPad ($120k for 7%) 13. The Mint ($500k for 10%) 14. AI2 Incubator ($50-$150k) 15. HF0 ($125k for 7% + $375k) 16. AI Grant ($250k Uncapped) 17. Betaworks AI Camp ($500k) 18. 500 Startups ($112.5k for 6%) 19. Entrepreneur First ($125k for 8%) 20. Y Combinator ($125k for 7% + $375k) 21. Conviction Embed ($150k Uncapped) 22. Founders Fellowship ($150k for 5-10%) 23. SPC Fellowship ($400k for 7% + $600k) (More info below) Who am I missing?
142
317
29
2,752
709,935
⚡️ Excited to share that I am starting an AI+Education company called Eureka Labs. The announcement: --- We are Eureka Labs and we are building a new kind of school that is AI native. How can we approach an ideal experience for learning something new? For example, in the case of physics one could imagine working through very high quality course materials together with Feynman, who is there to guide you every step of the way. Unfortunately, subject matter experts who are deeply passionate, great at teaching, infinitely patient and fluent in all of the world's languages are also very scarce and cannot personally tutor all 8 billion of us on demand. However, with recent progress in generative AI, this learning experience feels tractable. The teacher still designs the course materials, but they are supported, leveraged and scaled with an AI Teaching Assistant who is optimized to help guide the students through them. This Teacher + AI symbiosis could run an entire curriculum of courses on a common platform. If we are successful, it will be easy for anyone to learn anything, expanding education in both reach (a large number of people learning something) and extent (any one person learning a large amount of subjects, beyond what may be possible today unassisted). Our first product will be the world's obviously best AI course, LLM101n. This is an undergraduate-level class that guides the student through training their own AI, very similar to a smaller version of the AI Teaching Assistant itself. The course materials will be available online, but we also plan to run both digital and physical cohorts of people going through it together. Today, we are heads down building LLM101n, but we look forward to a future where AI is a key technology for increasing human potential. What would you like to learn? --- @EurekaLabsAI is the culmination of my passion in both AI and education over ~2 decades. My interest in education took me from YouTube tutorials on Rubik's cubes to starting CS231n at Stanford, to my more recent Zero-to-Hero AI series. While my work in AI took me from academic research at Stanford to real-world products at Tesla and AGI research at OpenAI. All of my work combining the two so far has only been part-time, as side quests to my "real job", so I am quite excited to dive in and build something great, professionally and full-time. It's still early days but I wanted to announce the company so that I can build publicly instead of keeping a secret that isn't. Outbound links with a bit more info in the reply!
1,494
3,552
1,073
27,449
2,553,017
The @CommonCrawl team just released all their statistics publicly ! And across all dumps, including the more recent one 🙌 Now is the time to finally analyze what's inside the source of most pre-training datasets out there 👀 1/n
2
23
2
119
14,504
👇👇 for biotech
we all want to and need to be prepared to train our own large-scale language models from scratch. why? 1. transparency or lack thereof 2. maintainability or lack thereof 3. compliance or lack thereof and because we can, thanks to amazing open-source and open-platform ecosystem. (1/12)
1
3
477
good and bad mistakes, shift to stretch mistakes for success. Interesting reading from my colleague Jason Katz: linkedin.com/posts/agabajer_…
1
2
149
“If you want to be successful, I would encourage you to grow a tolerance for failure.” ~ Nvidia CEO Jensen Huang
157
5,259
292
26,885
2,690,029
# on shortification of "learning" There are a lot of videos on YouTube/TikTok etc. that give the appearance of education, but if you look closely they are really just entertainment. This is very convenient for everyone involved : the people watching enjoy thinking they are learning (but actually they are just having fun). The people creating this content also enjoy it because fun has a much larger audience, fame and revenue. But as far as learning goes, this is a trap. This content is an epsilon away from watching the Bachelorette. It's like snacking on those "Garden Veggie Straws", which feel like you're eating healthy vegetables until you look at the ingredients. Learning is not supposed to be fun. It doesn't have to be actively not fun either, but the primary feeling should be that of effort. It should look a lot less like that "10 minute full body" workout from your local digital media creator and a lot more like a serious session at the gym. You want the mental equivalent of sweating. It's not that the quickie doesn't do anything, it's just that it is wildly suboptimal if you actually care to learn. I find it helpful to explicitly declare your intent up front as a sharp, binary variable in your mind. If you are consuming content: are you trying to be entertained or are you trying to learn? And if you are creating content: are you trying to entertain or are you trying to teach? You'll go down a different path in each case. Attempts to seek the stuff in between actually clamp to zero. So for those who actually want to learn. Unless you are trying to learn something narrow and specific, close those tabs with quick blog posts. Close those tabs of "Learn XYZ in 10 minutes". Consider the opportunity cost of snacking and seek the meal - the textbooks, docs, papers, manuals, longform. Allocate a 4 hour window. Don't just read, take notes, re-read, re-phrase, process, manipulate, learn. And for those actually trying to educate, please consider writing/recording longform, designed for someone to get "sweaty", especially in today's era of quantity over quality. Give someone a real workout. This is what I aspire to in my own educational work too. My audience will decrease. The ones that remain might not even like it. But at least we'll learn something.
658
3,619
837
18,756
2,297,572
Early thoughts on the Apple Vision Pro (I ended up buying directly in store last evening). I'm about 3 hours in, between late last night and this morning. The first major thing that must be said is WOW - the visual clarity is way beyond anything that came before. But, a bit unexpectedly, this is so in some strange mixed way - your surroundings (the passhtrough) are a bit blurry and even a tiny bit laggy. But anything rendered fully virtually, e.g. a screen is very sharp and easily readable. Super cool. I mean, just the simple experience of arranging a few windows around your living room and moving around them is incredible. I feel very creative thinking through and designing my ideal setup of all the apps in my space. Mind is blown and goes places. The second major thing is a bit less upbeat. This launch is not like the other Apple launches. It is off-brand. It is selectively and inconsistently either highly polished, or highly raw/undercooked, poorly throught through, janky or even straight up buggy. It's like some parts of the org get an A+ and some get an F. Or it's like some of them had 4 years to work on their part, and some had 4 months. It's like it was rushed a bit to "just ship" and basic UI/UX interactions weren't finished, thought-through or debugged. Jank Let me describe a bit some of the jank. The setup was a bit too long and janky for me. At one early point you're asked to bring your unlocked iPhone close, but you can't unlock your iPhone because your face is obviously covered so FaceID doesn't work... ?. Then I had some error connecting the phone to it so I had to go through "manual" setup. Then the sound wasn't working until I rebooted. Then I got an iMessage from a friend and I was shown a notification inside the Vision Pro about it, but when I clicked into iMessage app, it was fully empty - where is the message? When launching Guest Mode to show a friend, nothing tells you that you're supposed to also press the digital crown to activate. Very simple interactions are buggy - e.g. in the app store when I select an app to preview it and then hit back, I'm forced to for some reason go back 10 times through previously previewed apps to get back to the main screen, some bug or something. My Disney+ app never opened, it just spins forever, I'm not sure how to launch this app. When you launch Apple TV, there is zero indication or recognition of the fact that you're inside Vision Pro. No featured content, no custom content, no text indicating anything, no nothing. I'm not sure, I thought there would be a few surround videos or something? Also my brain: "$3500 for a Vision Pro? Yes two please! $9.99 for AppleTV+? Absolutely not." More generally, as you access Apple apps, a lot of them are just ignoring that you're inside a Vision Pro, and just pretending like nothing happened. I'd want new Spatial Content and interactions to be 100% front and center and featured. The "copy pasting" of stuff seems pervasive. The raw Spatial Computing OS is there, but it's almost like the OS is all there is. The apps that take advantage in any way of "Spatial Computing" seem few and are somehow also hard to find and/or not prominently featured. There's the little blue guy app who you can poke and he laughs. There's the jet engine app, which is kind of cool, but I wasn't actually really learning anything, it felt gimmicky, like an early demo. There are some really cool environments, but why are there only 5 of them?. There's what seems to be some early grifter content on the app store, from people trying to sell you e.g. a super basic looking watch app that just shows time, for $2.99. The ability to look at your laptop and just "connect" worked the second time, and it was glorious, wow. Your screen just shows up in your living room and you can use the keyboard/mouse. Very cool. The Vision Pro is sadly a little bit too heavy and it doesn't "disappear" due to this, even with the double strap (which is essential). I feel a bit pressure from the device on my head. But it's okay, we're at the edge of what is possible. A bunch of other small things. The world shakes a little bit with every step, especially if you land a bit harder on a heel. You have to unlearn and relearn some UIUX, because your eye gaze is now your active pointer. So you can't just look somewhere else a bit too early, before you "click" it. It's very cool that the eye tracking is so high quality. Anyway, I'm rambling. Conclusions. The hardware itself and the core Spatial Computing OS aspects exceed my expectations. I loved sprawling on my couch, opening up a few windows, and I half-watched a movie while scrolling through web. I loved pacing around my room arranging my digital work/entertainment space. I FaceTimed a friend and we laughed about how silly my digital avatar looks, haha. I pulled up Music and played the only thing I have in it - that U2 album that was given to everyone back in 2014. nice. I'm very happy with this early preview of what could be possible, and using the current experience as a prompt to explore it. Few recommendations to Apple come to mind: 1) eliminate simple bugs and jank. 2) fight early grifter content by featuring very very prominently any apps that are actually good, don't use dark patterns, are ideally free to try, and acknowledge in any way that the user is in a Vision Pro. 3) Consider a free subscription of AppleTV+, or maybe a $100 app gift card to those who purchase Vision Pro, so people don't lock up (?). It feels bad to pay that much money just to get in, and then immediately feeling like you're blocked behind additional pay walls, for experiences that could very well be very very raw and undercooked. 4) In general, feature a lot more prominently any content that is actually designed for spatial computing. I don't want to just put up iPad apps around me. I am simultaneously wearing a revolution in computing, and the software to actually show me around is not just absent but what is there is mildly janky and annoying. Ok, this concludes the section where I just "wing it" based on what I'm seeing, going in fairly blind, over the first ~3 hours. I will now do a bit more research, read more, watch some videos/tutorials, and come back for round 2.
227
396
117
5,646
1,727,247
A pediatrician who volunteered at a Gaza hospital describes what it's like there.
197
1,442
108
5,221
560,247
Had an interesting conversation with 11 yo when he said he worried a lot about his schoolwork. I told him the ambitious always have some worry, but you can't let yourself be consumed by it. Even if all you cared about was work, you do better work when you're happy.
35
107
14
1,861
216,372
In today’s #ISMBECCB2023 poster session, pass by #30 to chat with smiley 😊⁦@geertvangeest⁩ about #glittr glittr.org, ✨ a ⁦⁦@ISBSIB⁩ @SIBTraining⁩ #bioinformatics #training ressource 🌟
4
1
24
3,679