Building in AI + Legal Recruiting Tech South Park Commons, YC, and MIT Alum Serial product & tech focused founder Musings on technology and life
NYC
Joined January 2011
- Tweets820
- Following1.1K
- Followers614
- Likes3.2K
Love this. Purchased!
We raised $2M to build the screen-free interface for infinite children’s stories.
No screen, no mic, no camera.
Kids use physical blocks to co-create stories with AI. They don’t just listen, they get to build each story.
Blocks are heroes, cities, objects, rivals, or languages. Kids can even create their own.
For parents who want to encourage independent, creative play without a chatbot or endless feed.
For ages 3-8+.
$89. Shipping mid-November. Order today:
playgeni.com/products/playge…
0xByron.eth 🦇🔊 retweeted
Hi, I'm Felix. I'm an eng lead working on Claude. I love watching people take flight and do things they never thought possible.
We often describe that as "being more ambitious" when using AI, which is tricky advice to follow. I wrote a tutorial:
felixrieseberg.com/how-to-be…
Elon understands that most people will hire only one real "chief of staff" agent. The models are pretty much there so it's just a race to win on connectors, memory, context, and user habits. Model swapping can come later
This is a really big deal, mostly because of what it signals.
Some thoughts 👇
Obviously Elon is not happy that Opus 5.5 is the absolute best model right now (i'm sure he's riding the model team hard), but he accepts that there is no contest for Opus 5.5 rn.
The alternative strategy is to stubbornly keep vertically integrating. This is still *obviously* the best long term strategy, so this indicates his belief in this product category. Worth the short term adjustment.
That tells you something about i) the growth numbers of grok bot and ii) his belief that winning early matters for this market.
Competition is high but grok bot can stay ahead. OpenAI botched their Dots launch, Anthropic doesn't have one yet, and Instinct is dangerous with the new funding (they could win and then be bought by openAI).
Elon is pissed he's behind in AI, and this is the first time he has a best-in-ai-category product. As I've shared before, i think Grok bot is easily the best in their category, the cursor team cooked hard.
Unlocking Suno and Midjourney is very cool. These don't have official API products! He just went and got those deals done (they will make the product much better).
Even though this sounds like "we're neutral now", to me it seems that he's hyper-focused on beating openAI. I am guessing this personal.
I don't think many founders are able to make a choice like this. It's extremely Elon coded. Can you imagine OpenAI or Ant doing something like this? Impossible imo.
Codex tip for multi account users: after a non-banked reset, send a quick message in every account to start the 7 day clock. Chatgpt Work in browser can be easier for other accounts you're not signed into the app.
0xByron.eth 🦇🔊 retweeted
>be me
>discover effective altruism
>apparently normal charity is inefficient
>why donate to random sad thing when spreadsheet can tell you optimal sad thing
>fair enough
>buy mosquito nets
>save lives
>numbers look good
>feel powerful
>couple years later
>someone asks an innocent question
>why only count people alive today
>huh
>future people matter too
>obviously
>my grandchildren shouldn't matter less just because they haven't spawned yet
>reasonable.jpg
>keep following logic
>what about their grandchildren
>also yes
>what about people in 500 years
>sure
>5000 years
>why not
>500 million years
>starting to get weird but morality is morality
>open calculator
>humanity could survive for an astronomically long time
>could colonize galaxy
>could have trillions upon trillions of descendants
>maybe digital people too
>maybe simulated civilizations
>maybe dyson spheres full of happy uploaded minds
>calculator starts smoking
>realize currently living humans are rounding error
>8 billion people suddenly looking extremely beta
>future contains potentially 10^something people
>can't even fit beneficiaries in google sheets
>new moral priority unlocked
>protect the long-term future
>stop thinking in units of "people helped"
>start thinking in "fraction of cosmic endowment preserved"
>malaria?
>terrible
>but only kills existing humans
>AI extinction could delete the entire light cone
>nuclear war could permanently derail civilization
>bad institutions could lock in terrible values for ten million years
>someone invents wrong constitution in 2140
>quadrillions suffer
>better fund governance workshop now
>friend says maybe we should improve hospitals
>explain opportunity cost
>friend says hospitals are full of actual sick people
>explain scope sensitivity
>friend stops inviting me to dinner
>need to decide what to fund
>easy
>expected value
>suppose project has one in a million chance of preventing extinction
>sounds tiny
>but extinction destroys 10^50 future lives
>multiply
>mother of god
>$10 million project has expected value of several galaxies
>charity evaluation complete
>someone asks where the one-in-a-million number came from
>expert judgement
>which expert
>us
>how calibrated
>extremely thoughtfully
>reduce estimate to one in ten million to be conservative
>still beats curing cancer by 38 orders of magnitude
>epistemic robustness achieved
>someone says maybe project doesn't work
>assign 20% chance
>still astronomical
>maybe project makes problem worse
>assign 5% chance
>still astronomical
>why 5
>because 30 felt pessimistic
>publish 46-page report
>contains seventeen sensitivity analyses
>every sensitivity analysis begins after assuming intervention has positive sign
>critic says you're multiplying enormous hypothetical stakes by extremely uncertain probabilities
>yes
>that's literally why it's important
>critic says the uncertainty might be structural rather than numerical
>make probability smaller
>critic says no, I mean maybe your model is wrong
>make probability smaller again
>critic begins rubbing temples
>discover AI safety
>perfect longtermist cause
>AI might kill everyone
>or create utopia
>or seize galaxy
>or tile universe with paperclips
>or create billions of conscious software minds
>finally a problem with numbers big enough for me
>start AI safety nonprofit
>mission: prevent dangerous AI
>hire smartest people available
>smartest people immediately start building better AI to understand dangerous AI
>interesting
>we must understand capabilities to understand safety
>we must scale models to study alignment
>we must race ahead so less responsible actors don't get there first
>we must deploy systems to learn how deployment can go wrong
>we must build the thing quickly because building the thing quickly is dangerous
>outsider asks why the people most worried about AI apocalypse all work at AI companies
>complicated field
>company releases stronger model
>very concerned
>company begins training even stronger model
>extremely concerned
>company raises $14 billion
>concern reaches unprecedented levels
>need to influence government
>future is at stake
>normal democratic process too slow
>politicians don't understand exponential curves
>public doesn't understand x-risk
>experts must guide them
>who counts as expert
>people who understand x-risk
>who understands x-risk
>our friends
>someone objects that this seems politically convenient
>explain we're representing future generations
>future generations unavailable for comment
>develop concept of value lock-in
>terrifying possibility that one ideology controls civilization forever
>therefore extremely important that civilization adopts correct values before lock-in
>whose values
>let's circle back
>begin with impartial morality
>end with small group of people deciding what quadrillions of hypothetical beings would want
>beautiful arc
>meanwhile actual humans keep doing annoying things
>voting wrong
>having parochial attachments
>loving family more than strangers
>caring about local community
>getting upset when told their suffering is cosmically negligible
>evolutionary biases everywhere
>explain that moral intuition cannot be trusted
>except intuition that future digital people count
>and intuition that extinction is uniquely bad
>and intuition that our probability estimates are sane
>and intuition that our institutional choices improve the future
>those intuitions survived peer review
>someone donates $5k to local homeless shelter
>inefficient
>could have funded 0.0000000000003% of an AI governance researcher
>think of all the simulated people you just killed
>okay maybe don't phrase it that way publicly
>PR team says "future generations deserve a voice"
>much better
>journalist asks what longtermism means
>say "future people matter"
>everyone agrees
>great
>journalist asks what follows from that
>well technically we should redirect enormous resources toward low-probability interventions affecting astronomical futures
>journalist raises eyebrow
>return to "future people matter"
>motte has entered the chat
>critic: of course future people matter
>me: glad we agree
>critic: I don't agree that your institute knows how to help them
>me: why do you hate our grandchildren
>eventually notice uncomfortable implication
>if future value dominates everything
>then helping people today mostly matters through effects on future
>education matters because future institutions
>health matters because future productivity
>democracy matters because future trajectory
>human beings slowly become instrumental variables in their own moral philosophy
>see starving child
>feel compassion
>check spreadsheet
>child's direct welfare contribution negligible
>but perhaps childhood nutrition improves national institutional quality
>compassion restored
>tell myself this is impartial altruism
>one day assistant asks obvious question
>"how do you know your intervention actually improves the far future?"
>silence
>open spreadsheet
>increase column width
>add confidence interval
>assistant asks again
>"no, I mean how do you know the sign is positive?"
>stare into cosmic light cone
>10^50 people staring back
>none of them exist
>none of them can tell me
>none of them can falsify my assumptions
>realize I have invented the perfect constituency
>infinitely important
>completely silent
>and always represented by me
Ideal setup for budget tokenmaxxing. Cloud agents are great but $$. Instead:
- buy used SFF PCs. Dell Optiplex i7-12700T w 32GB ram is a good value currently (~$500)
- install linux, put in basement
- SSH machines for coding (Orca / Herdr)
- self-host Github Runners for CI/CD pipeline (instead of Github Actions)
- Trigger.dev self hosted for background jobs
- create skills that document your full setup. new machines can be then spun up in a single command.
- create custom TUI apps to monitor machine health (mine attached)
- create scripts to automate common actions on fleet (pull git, update packages, etc)
0xByron.eth 🦇🔊 retweeted
Our new agent generates kitchen plans in 30 seconds
…with editable construction docs:
- elevations & floorplans
- realistic 3D
- live quotes (real manufacturer SKU's & prices)
We're solving kitchens first because it's the most difficult remodel, and will unlock the whole home.
For architecture work, humans and agents will need to collaboratively prototype in one place… We believe the solution that wins long-term will consume the least cost (i.e. mental strain, time, tokens/energy).
-> In this video, I drop a google image, 80% of the plans get built by the agent, and then I make the last batch of edits manually
This is possible because:
- in 1 year, we built a procedural 3D engine fully from scratch (ThreeJS x WebGPU)
- The agent only determines the high-level parameters, delegating the low-level "physics" stuff to the engine
I go more in-depth on what we’re building here:
nikstakheiko.substack.com/p/…
On PG vs Elon:
PG is a reasoning model that does well when thinking budget is xhigh+. His essays and initial posts are meticulously reviewed and flawless. But off-the-cuff replies can be iffy.
Elon has big model smell. Reasoning low when X posting, or giving live interviews. Turns up to xhigh when needed.
It rarely makes sense to get a super powerful laptop for dev. Remote dev machine is better for cost, battery, always on, weight, upgrade cycle, flexibility. Unless you're flying a lot without starlink, fat-server-thin-laptop is the way.
Field notes for new macs:
TLDR I think the M5 Max is gonna work better for a lot of home maxxers
32GB M6: comfortably runs Gemma 4 12B, Gemma 4 26B-A4B, and Qwen3.8-27B at 4-bit. Even Gemma 4 31B Q4 is only ~17.4GB of weights, although long context/KV cache makes 32GB cramped.
64GB M5 Max: the sensible floor for serious local AI: Qwen3.8-27B, Gemma 4 31B, Qwen3.6-35B-A3B with lots of working/context headroom.
128GB M5 Max: qualitatively different. Qwen3.5-122B-A10B 4-bit uses ~67GB and has already benchmarked at roughly 64–65 tok/s on M5 Max.
256GB M5 Ultra: now you’re targeting monsters like DeepSeek-V4-Flash: 284B total / 13B active.
512GB M5 Ultra: useful for enormous frontier MoEs, but Kimi K3 is 2.8T parameters / 104B active—roughly 1.4TB of weights at 4-bit, so the interesting claim is really multi-Mac distributed inference, not one Studio magically running K3.
The value of your work is roughly the difference between your work output and what can be achieved using a simple prompt
Grok Bot is such a delightful experience. In theory you can do everything it does with Hermes / Openclaw, but it just takes away all the annoying friction that makes those painful to set up and use.
Yet they kept the power of a full, always-on machine. For example, my Grok Bot machine has been added to my Tailscale network and can SSH back and forth - it's now part of my cloud fleet.
Plugins are so easy. Really kudos to the @cursor_ai / @xai product teams on this one.
It makes setting up bots so damn fun!
Ask your agents to audit your session logs for which MCPs are used, and which tools were actually used for them. Then, it can disable all tools that you haven't used in the past 6 months.
Save cost and intelligence, re-enable the tools later if you need them.
For apps that are UI controlled (ie Grok Bot, you can ask Codex computer-use to handle it)
After going back and forth between in-repo .md files vs cloud-hosted task management, I'm now firmly a believer in cloud tasks. Currently using Linear, wayy better for running agents across multiple machines without worrying about syncing task files.
In theory you lose some history and searchability, but I haven't found that to be impactful.
Moved a chunk of my github actions (Vitest) to a self-hosted action runner. Was spending $10 / day, moved to an old linux machine in just a couple prompts. Super simple, recommend for anyone wasting $$ on actions.
chatGPT Pro connected to github is pretty crazy. It has access to your codebase, can do coding and even run simple tests and create branches/PRs. I’m now giving it my hardest tasks for planning and architecture review; it’s even better than fable.
It is slow but super thorough. And doesn’t use codex limits!
Imagine if your cell service was
- reliable everywhere, no dead zones
- worked globally with one plan
- high / unlimited data
- cheap
Yeah I don't think cell service is solved.
Starlink was / is great because it leapfrogged a service that was woefully terrible: satellite internet
Conventional sat internet (HughesNet, Viasat) and other last-mile internet services were PAINFULLY bad before Starlink made them irrelevant.
But... that's *not* the case for current cell service.
The reality is, with few exceptions, existing 4g / 5g cell coverage is very usable or even great in most populated parts of the U.S. and all modern countries.
The speeds are good, the plans are good, there's lots of competition and options for consumers to choose from.
I pay like $35 / month for unlimited 5g and have not wondered about switching providers in many years.
In other words, the market conditions that allowed Starlink to succeed (solving a real pain-point) simply do not exist for everyday phone use.
This is not anti-Starlink at all (I was an early Dishy beta user and now have a Mini) but an observation.
Am I wrong?
0xByron.eth 🦇🔊 retweeted
Contrast two kinds of men now arguing over America's future.
One looked at empty Texas scrubland and saw the largest building on Earth, then summoned $16.8 billion in capital, thousands of construction tradesmen, and 3,000 permanent minds to raise and run it. Before that he saw rockets landing upright when every expert called it fantasy, and electric cars people would line up to buy, and he materialized those too. His tools are reason, vision, and trade. Every dollar he commands was handed to him by a customer who got something better in return.
The other kind is the politician who could not pour one yard of that concrete. But he holds the one tool of his trade, the state's gun. He calls the building excess, drafts the wealth tax to seize its output, and names the seizure compassion and justice. Ask the senator demanding Musk's billions what he has ever built. His one skill is the forced transfer of what abler men have made.
When Terafab rises, understand what you are looking at. The unforced human mind, made visible at scale. The looters will still demand their cut, and the demand itself is the confession. Nobody loots an empty desert. Wealth must be created before it can be seized, and it is created by men like Musk.
