Cisoml

@benjitaylor Let us use our bookmarks as context for our grok research

First thing on the agenda: improve everything

@amitisinvesting @imagine

SPY loses $650 time to bust out the peter lynch clips just exhausting i will say, the covered call has helped this year if you aren't actively transitioning your entire portfolio into new themes every week for the people buying calls on the idea that we will break out...all i can say is thank you for letting me sell them to you because id rather be stuck having to roll a position if we do get some huge V shape vs not getting SOME premium from the market when we chop every single week

Shield AI just completed its latest round at $12.7B. It has been such an incredible 10 years with an amazing team. Grateful to so many for the opportunity to contribute to noble mission and the trust placed in us by warfighters around the world.

Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual difference. I am mildly surprised that my very first naive attempt already worked this well on top of what I thought was already a fairly manually well-tuned project. This is a first for me because I am very used to doing the iterative optimization of neural network training manually. You come up with ideas, you implement them, you check if they work (better validation loss), you come up with new ideas based on that, you read some papers for inspiration, etc etc. This is the bread and butter of what I do daily for 2 decades. Seeing the agent do this entire workflow end-to-end and all by itself as it worked through approx. 700 changes autonomously is wild. It really looked at the sequence of results of experiments and used that to plan the next ones. It's not novel, ground-breaking "research" (yet), but all the adjustments are "real", I didn't find them manually previously, and they stack up and actually improved nanochat. Among the bigger things e.g.: - It noticed an oversight that my parameterless QKnorm didn't have a scaler multiplier attached, so my attention was too diffuse. The agent found multipliers to sharpen it, pointing to future work. - It found that the Value Embeddings really like regularization and I wasn't applying any (oops). - It found that my banded attention was too conservative (i forgot to tune it). - It found that AdamW betas were all messed up. - It tuned the weight decay schedule. - It tuned the network initialization. This is on top of all the tuning I've already done over a good amount of time. The exact commit is here, from this "round 1" of autoresearch. I am going to kick off "round 2", and in parallel I am looking at how multiple agents can collaborate to unlock parallelism. github.com/karpathy/nanoc… All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale of course - you don't just have a single train. py file to tune. But doing it is "just engineering" and it's going to work. You spin up a swarm of agents, you have them collaborate to tune smaller models, you promote the most promising ideas to increasingly larger scales, and humans (optionally) contribute on the edges. And more generally, *any* metric you care about that is reasonably efficient to evaluate (or that has more efficient proxy metrics such as training a smaller network) can be autoresearched by an agent swarm. It's worth thinking about whether your problem falls into this bucket too.

@Starlink @OCCUPYMARS2025 When will @Tesla ‘s car be able to use it?

Starlink Mobile’s next-gen satellites will deliver 5G speeds from space with 100x the data density of the current V1 generation satellites V2 satellites will seamlessly enable streaming, internet browsing, high-speed apps and voice calls, just like being connected to a terrestrial network → starlink.com/mobile

@elonmusk It would be great if @Tesla ‘s sentry mode had @grok notifications that would alert us when someone damaged our car!

Grok 4.20

@elonmusk @cgarciae88 @pbeisel It would also be great if our comments after disconnecting FSD would persist in its memory (lots of edge cases like gated community lane selection, for example)

@cgarciae88 @pbeisel Coming

I "drove" to a meeting this morning. Actually I didn't drive at all, FSD did. From my driveway to the parking spot, 30 miles. In fact, I barely remember the drive, my brain was very focused on something else. On FSD v14: 85% of my interventions: finding a parking spot. 15% wanting to take a different route. 0% safety critical. 100% true statement.

First Cybercab off the production line at Giga Texas

xAI’s EPIC 'All Hands' Update Today Elon & team talk Grok, X, SpaceX ► Silences removed (to save you time) ► Boosted audio (for easier listening) Timestamps: 0:00 - Elon Musk’s Opening Remarks - xAI Accomplishments Since Inception 3:58 - Elon & xAI Team Give Update 26:00 - Live Tour Of xAI’s ‘Macrohard’ AI Training Supercluster In Memphis — It’s INSANE 30:20 - xAI’s Secret Weapon: The X Platform - Nikita Explains 32:58 - Elon On X Money, X Chat, Future Goals 35:34 - Elon Explains SpaceX & xAI Joining - “Exploring The Universe” & Moonbase Alpha 38:43 - My Recap & Key Takeaways

Building an interstellar civilization #xai-joins-spacex" target="_blank" rel="nofollow noopener">spacex.com/updates#xai-jo…

SpaceX has acquired xAI, forming one of the most ambitious, vertically integrated innovation engines on (and off) Earth → #xai-joins-spacex" target="_blank" rel="nofollow noopener">spacex.com/updates#xai-jo…

@PalmerLuckey @anduriltech Any news on delivery for the ones ordered in December?

Alroght, as promised, we have re-opened orders for the Anduril Chromatic. The original production run sold out in just a few minutes, this final preorder closes at midnight on Sunday. They are a gigantic pain in the butt to make, no amount of pleading will bring it back!

@Osinttechnical At what cost?

The small French manufacturer ALM Meca just came out of nowhere to announce Fury, the company’s privately developed, low cost interceptor. The drone is capable of hitting speeds of over 700 km/h and pulling 20 G, giving it massive engagement capabilities.

@recap_david ASSISTANT

$24,000 per year from this simple AI Dentist Voice Agent (and why I’m probably insane for giving it away for free) One dental practice was bleeding $6,000+ every month from missed after-hours phone calls. That’s 20–25 potential new patients disappearing simply because no one was there to answer and book. So I built an AI voice receptionist that books dental appointments 24/7, powered by n8n + ElevenLabs, and guided by the clinic’s internal rules and real scheduling availability. Here’s what the system handles automatically: → Picks up calls with a natural-sounding AI receptionist → Captures patient info + insurance details → Checks live calendar availability → Schedules appointments instantly → Logs every interaction into a Google Sheet The wild part? A nearly identical AI voice system was sold to a dental practice for $24,000 per year by another founder. And this isn’t limited to dentists. Any service business losing revenue from missed calls can deploy the exact same setup. Want the full n8n workflow template? 1. Retweet & Like 2. Follow me 3. Comment “ASSISTANT” I’ll send you the entire system for free, plus a step-by-step setup video — including all the ElevenLabs automation pieces. (must be following so I can dm you)

@amitisinvesting @a16z Congrats Amit! I am sure you will excel on this new path too!!

small life update! i’ve joined Andreessen Horowitz’s @a16z venture scout program! they are essentially letting me manage a little bit of money to write checks into startups. super grateful for the opportunity and excited to begin my angel investing career with the institutional backing of a firm as storied as Andreessen Horowitz (early into facebook, twitter, airbnb, lyft, slack, coinbase and many more) big thank you to @dhaber for the connection and opportunity i’ve had a little success with investing in public equities, I have no idea if any of those skills could translate to the private markets. investing at pre-seed/seed requires a different paradigm as there are no numbers, financials, track-record, and quite frankly there may not even be a product – it really is a bet on the founder and their vision. having said that, if you are building something really cool and raising money, I’d love to hear about it. please email amit@akcomms.com. you can include anything you have whether that is a deck, one pager, etc. i’ve also built out a website that anyone can read to get a sense of how i think of investing in general and the 5 startups I’ve invested in over the past year (one has 20x’ed, one has 10x’ed, the other 3 just finished their seed round last month) no clue if I will be good at this but I appreciate a16z for the opportunity and hopefully we can try to find the next uber or robinhood being built in the early days – excited to see any pitches and please don’t hesitate to reach out! angel investing philosophy below!

@elonmusk Starship 4 > 420 feet?😜

Starship 4 will be another 10% to 20% longer