OmarH

@johnarnold The first thing people do now is ask an LLM. That alone might cut down on unnecessary reversals. If even a few patients push back with better-informed questions, some “archaic” doctors may finally start second-guessing themselves.

Many years ago I hurt my knee playing sports. I was referred to the orthopedist for one of the local pro teams. After keeping me waiting for 2.5 hours, he diagnosed a cartilage tear and recommended surgery. I was so mad at his manner and tardiness I left without scheduling. The next week I got a second opinion from a much younger doc who was likely more current on the recent medical literature. He looked at the same MRI. He said he could do surgery now but his advice was to wait 30 days and see if it healed on its own. It did. Medical reversal is when a practice that became widely used is later shown to be ineffective or even harmful. Examples like meniscus surgery show the need to keep gathering evidence. A not immaterial part of the practice of modern medicine doesn't improve health, and may be net harmful.

@ClementDelangue @Pinterest @Airbnb @NotionHQ @cursor_ai @eoghan @intercom This is why I’m bullish platforms like @getmixlayer where you can lean into the advantages of accessing the weights and not just compete on token price.

After @Pinterest @Airbnb @NotionHQ @cursor_ai, today it’s @eoghan @intercom publicly sharing that they’re finding it better, cheaper, faster to use and train open models themselves rather than use APIs for many tasks. And hundreds of other companies are doing the same without sharing. Ultimately, I believe the majority of AI workflows will be in-house based on open-source (vs API). It took much more time than we anticipated but it’s happening now!

@TexasVC @_ZachGriff @OneMileataTime @evan7257 Trying to convince my spouse to have another baby so we can take advantage of this.

If you're flying out of IAH and have a United boarding pass with 1K or GS status, there is a special security line at Terminal C with no wait. They are also allowing a parent with a carried infant (NOT 2 year olds). @_ZachGriff @OneMileataTime @evan7257

@scottastevenson @rabois I think you can probably work something out if you discuss it. People typically won’t sit and work out all the possible variations of every position they will or won’t accept. (For instance when you trade stocks, you don’t price for positions deep in/out of the money.)

@realohtweets @rabois Maybe you provide fallback positions. But if this is unacceptable, maybe the agent goes to get proposals from other vendors in order to get leverage or an alternative path.

It’s inevitable that contracts will be negotiated by agents within the next decade. Business will speed up so much that humans won’t be able to keep up with day-to-day contracts between companies. Lawyers will set the policies and audit— agents will be trusted to do the rest. This sounds irresponsible, until you realize what the error rate is for human contract managers—fairly high. I believe agents will be a safer way to negotiate rudimentary agreements within the decade, just like self-driving is now safer than human driving. This chart will exist for basic business contracts:

1/ Today, I'm excited to announce ✨Mixlayer✨. Two years ago, I reoriented my life around a singular mission: to build the best inference platform for open source AI.

@zackangelo @JustinLin610 Agree it’s the a good bet to boost value, but salvaging what?

Alibaba's best bet for salvaging this is to try and make this into a spin out, fund @JustinLin610 and his team with full autonomy and bring them back in after they have some traction

@buccocapital @RyeNotBerben Why don’t all the LLMs start a search function (for research) and have their search bot do the crawl? Or do people permit the search crawl only for certain companies?

@RyeNotBerben Yes. A meaningful plank in the thesis x.com/buccocapital/s…

Sounds like $googl won and others are screwed, no?

@DrSyedHaider @CharlesFLehman I was going to say and real estate in talent cluster cities but this is a better answer

@CharlesFLehman Doubtful. This will be used to herd everyone towards online digital ID.

A friend observed a while back that if you're bullish on AI, the best investment is airline stocks.

Trying to reconcile the "Claude did in 1h what took a Google team a year" thing with my own LLM experiences. It creates lots of code for me, which helps, but also that code leaks resources, is prone to data races, and maintainability is an afterthought. Where is the disconnect?

@ohabryka The way forward is to let the AI build this massive, magnificent piece of code and then figure out what it did right or wrong in little pieces on the backend (perhaps using cursor) rather than throttle it on the front end.

@ohabryka I am where you are: much prefer the ide navigation and comprehension. I think where people have shifted to the CLI, it’s because they’ve given up trying to grok the whole codebase. The AI has blown past our puny human minds and we are just holding it back by catching up.

Ok, can someone explain to me why people use Claude Code instead of Cursor? I get obvious productivity boosts from having my IDE integrated with my AI tooling, and lose that in Claude Code, but everyone keeps raving on about how great Claude Code is, so what am I missing?

@vkhosla @grok please outline in text form

One of our most requested resources within our venture assistance is help with pitching and storytelling -- here's the workshop on "Nailing your Fundraise" i gave our CEOs at our summit and full deck linked on our site khoslaventures.com/nail-your-rais…

@bp22 @ChatGPTapp @claudeai It can’t directly make the PDFs. It sort of has to write python code that then generates the PDF. And it’s very rare that anyone has to do that in the wild so there are few good examples. Easier to have the LLM write it to markdown (just text) and have pandoc make the PDF.

Why does @ChatGPTapp and @claudeai go in full idiot mode when you ask it to put info into a pdf?

@johnarnold This is an older thread but a new argument dropped x.com/levie/status/2…

@realohtweets maybe. I guess there's a Laffer curve equivalence where there is a peak employment and question is what side are we on.

It's hard to know how much of the collapse in software developer job listings is a reduction in job hopping vs fewer total jobs. That said, there's a real disconnect between # of jobs open and record # of comp sci college grads, and that's before AI coding tools get very good.

@aakashgupta A similar thesis was shared with me by @ChrisCamillo a few months ago.

The entire robotics industry is about to compress a decade of progress into 18 months, and nobody’s pricing it in. The hardware has been ready for years. Boston Dynamics had Atlas doing backflips in 2018. The bottleneck was never motors or actuators. It was that every robot behavior had to be hand-coded. Pick up a box? That’s one program. Pick up a bottle? Different program. Move the box from shelf A to shelf B in a warehouse with slightly different lighting? Start over. Foundation models broke this completely. Before VLAs, teaching a robot one skill gave you exactly one skill. Zero compounding. Zero transfer. A robot trained to fold shirts couldn’t fold towels without starting from scratch. The labor intensity of data generation meant robotics datasets stayed narrow, robots overfit, and small variations like object weight or table height caused failures. Now a single Gemini Robotics model handles tasks it has never seen in training. Google’s On-Device model learns new behaviors with 50-100 demonstrations. Not 50,000. Fifty. That’s a 1000x reduction in the data requirement for new capabilities. The speed implications cascade through everything. First order: deployment timelines collapse. What took robotics teams 6-12 months of custom programming now takes days of fine-tuning. Second order: the addressable market explodes. Tasks that were never economical to automate suddenly are, because the integration cost dropped by orders of magnitude. Third order: the data flywheel accelerates. Every robot running Gemini Robotics feeds learning back into the foundation model. More deployments means faster improvement means more deployments. Physical Intelligence raised at $2.4B because investors finally understood this. Boston Dynamics partnered with Toyota Research Institute to bolt Large Behavior Models onto Atlas. Every humanoid company is scrambling to either build or license the intelligence layer they don’t have. The market is still valuing robotics companies on their hardware differentiation. But hardware is commoditizing. Boston Dynamics spent a decade perfecting locomotion, and now that’s table stakes. The value is migrating entirely to whoever owns the foundation model that generalizes across embodiments. Google trained Gemini on the largest multimodal corpus ever assembled. Then they added physical actions as an output modality. That’s not a robotics company bolting on AI. That’s an AI company whose models now output motor commands. The companies pricing this correctly are building around foundation model access, not around proprietary hardware. The companies pricing this wrong are still acting like the moat is in the mechanical engineering. AGI moving into the physical world isn’t a 10-year prediction. Gemini Robotics shipped in March. The 1.5 version with chain-of-thought reasoning shipped in September. They’re iterating on a 6-month release cycle while hardware companies iterate on 3-year cycles. The gap between software intelligence timelines and hardware development timelines is the entire trade.

@rasbt @karpathy Flipped classroom, right. Flipped classroom = lecture at home, assignment in the class

@karpathy Yes, I guess what would make the most sense is a) Learn at home, use AI as a teaching tool to ask questions about the material, have it prep and test you, etc. b) Come to the classroom to do your homework, exams, etc.

A number of people are talking about implications of AI to schools. I spoke about some of my thoughts to a school board earlier, some highlights: 1. You will never be able to detect the use of AI in homework. Full stop. All "detectors" of AI imo don't really work, can be defeated in various ways, and are in principle doomed to fail. You have to assume that any work done outside classroom has used AI. 2. Therefore, the majority of grading has to shift to in-class work (instead of at-home assignments), in settings where teachers can physically monitor students. The students remain motivated to learn how to solve problems without AI because they know they will be evaluated without it in class later. 3. We want students to be able to use AI, it is here to stay and it is extremely powerful, but we also don't want students to be naked in the world without it. Using the calculator as an example of a historically disruptive technology, school teaches you how to do all the basic math & arithmetic so that you can in principle do it by hand, even if calculators are pervasive and greatly speed up work in practical settings. In addition, you understand what it's doing for you, so should it give you a wrong answer (e.g. you mistyped "prompt"), you should be able to notice it, gut check it, verify it in some other way, etc. The verification ability is especially important in the case of AI, which is presently a lot more fallible in a great variety of ways compared to calculators. 4. A lot of the evaluation settings remain at teacher's discretion and involve a creative design space of no tools, cheatsheets, open book, provided AI responses, direct internet/AI access, etc. TLDR the goal is that the students are proficient in the use of AI, but can also exist without it, and imo the only way to get there is to flip classes around and move the majority of testing to in class settings.

@zackangelo This is the opposite of what I thought you were saying and makes more sense.

@realohtweets Yep, by default ChatGPT will bring in information it's learned about you from your previous chats. It's rarely useful imo. I'd like to disable just that part without having to resort to a full blown temporary chat.

really wish there was an option to have ChatGPT ignore memory but stay in history. a lot of times it'll inject things from memory that aren't really useful.

@zackangelo Ah! Got it.

@realohtweets But doesn't that do that for all of your chats? I'd like to be able to turn it on and off for each chat.