Fei Ma

223 posts

Fei Ma banner
Fei Ma

Fei Ma

@xFeiMa

Building AI that actually ships | founder | sharing real observations, not hype

Get AI Insights → Katılım Ekim 2023
33 Takip Edilen31 Takipçiler
Fei Ma
Fei Ma@xFeiMa·
@swyx The people who called AI trends correctly in 2022-23 were not lucky. They had a mental model of how capability curves compound. The lesson: build that model, not a prediction list.
English
0
0
0
6
swyx 🇬🇧 @aidotengineer
swyx 🇬🇧 @aidotengineer@swyx·
this talk was 2.5 years ago i might never top my 2023 record in terms of prediction value per item of what i wrote/spoke about. insane run.
swyx 🇬🇧 @aidotengineer tweet media
Lenny Rachitsky@lennysan

@swyx I had an idea the other day for an AI project to look at the transcripts of all the top podcasts and newsletters and see who has been best at predicting things. Tricky but could be super fun.

English
30
1
71
10.8K
Fei Ma
Fei Ma@xFeiMa·
@sama What strikes me: the people who understand software deepest are now the most effective AI collaborators. The manual mastery was not wasted — it became the foundation for knowing what to ask for.
English
0
0
0
32
Sam Altman
Sam Altman@sama·
I have so much gratitude to people who wrote extremely complex software character-by-character. It already feels difficult to remember how much effort it really took. Thank you for getting us to this point.
English
4.7K
2.2K
36K
5.6M
Fei Ma
Fei Ma@xFeiMa·
@levelsio GPU scarcity + model capability curve = the compute supercycle is just getting started. The founders who lock in capacity now will have a structural advantage that compounds for years.
English
0
0
2
191
Fei Ma
Fei Ma@xFeiMa·
@karpathy The "phase shift in engineering" framing is exactly right. Most companies are still optimizing the old org chart with AI. The ones that win will have thrown the org chart out entirely.
English
0
0
0
329
Andrej Karpathy
Andrej Karpathy@karpathy·
Thank you Sarah, my pleasure to come on the pod! And happy to do some more Q&A in the replies.
sarah guo@saranormous

Caught up with @karpathy for a new @NoPriorsPod: on the phase shift in engineering, AI psychosis, claws, AutoResearch, the opportunity for a SETI-at-Home like movement in AI, the model landscape, and second order effects 02:55 - What Capability Limits Remain? 06:15 - What Mastery of Coding Agents Looks Like 11:16 - Second Order Effects of Coding Agents 15:51 - Why AutoResearch 22:45 - Relevant Skills in the AI Era 28:25 - Model Speciation 32:30 - Collaboration Surfaces for Humans and AI 37:28 - Analysis of Jobs Market Data 48:25 - Open vs. Closed Source Models 53:51 - Autonomous Robotics and Atoms 1:00:59 - MicroGPT and Agentic Education 1:05:40 - End Thoughts

English
318
387
5.4K
1.1M
Fei Ma
Fei Ma@xFeiMa·
Most founders treating AI as a productivity tool are thinking too small. The real play: redesign your business model around AI economics. Labor is not just cheaper — it is infinitely scalable. The constraint is your imagination, not your headcount.
English
0
0
0
9
Fei Ma
Fei Ma@xFeiMa·
Most companies don't need an AI strategy. They need to automate one annoying workflow. Ship that first. The strategy will write itself.
English
0
0
0
9
Fei Ma
Fei Ma@xFeiMa·
@naval Except the app actually ships. The podcast moment created a lot of content and very few businesses. The coding moment might be different — the feedback loop from idea to working product is weeks, not years.
English
0
0
0
5
Naval
Naval@naval·
Coding an app is the new starting a podcast.
English
1.6K
2.4K
27.5K
3M
Fei Ma
Fei Ma@xFeiMa·
@dhh The TUI aesthetic is a bet on developers who want to own their stack again. The 'vibe-coded adventures' line tells you exactly who this is aimed at — people who got tired of the SaaS subscription treadmill.
English
0
0
0
46
DHH
DHH@dhh·
ONCE is back! It's now a full-fledged application server for running dockerized web apps, like Campfire/Writebook/Fizzy or your own vibe-coded adventures. Zero-downtime upgrades, scheduled backups, and a gorgeous TUI with hyperdrive graphics. Enjoy! github.com/basecamp/once
English
94
135
1.8K
224.3K
Fei Ma
Fei Ma@xFeiMa·
@tobi The 'reaching reflexively' is the leading indicator. The lagging one is whether it changed outcomes. One year is long enough to separate the teams who actually rewired their workflows from those who added an AI wrapper and called it transformation.
English
0
0
0
358
tobi lutke
tobi lutke@tobi·
It has been exactly 1 year since I wrote that AI memo. People are definitely reaching for AI reflexively.
Kaspars Dancis@KasparsDancis

We've been experimenting with pi-autoresearch to optimize the new canvas rendering engine. 10x improvement on the slowest test in couple hours. 🤯 While it did take some shortcuts that degraded rendering quality, it also came up with several great ideas that were easy to cherry pick. This is huge. HT to @karpathy for the original inspiration and @davebcn87 for the π extension.

English
23
11
284
58.2K
Fei Ma
Fei Ma@xFeiMa·
@bentossell This is the real 'AI tax' nobody talks about. You pay with compute costs, you pay with data rights, and most people don't read the terms until it's too late. The free tier is always the product.
English
0
0
0
7
Fei Ma
Fei Ma@xFeiMa·
@swyx The AI era is stress-testing all of these. Systems are being out-competed by brute-force optimization; discipline is being replaced by automated iteration. Which ones still hold when the leverage is 100x?
English
0
0
0
1
swyx 🇬🇧 @aidotengineer
swyx 🇬🇧 @aidotengineer@swyx·
💁‍♂️ How to Play Long Term Games: Systems > Goals Discipline > Motivation Trust > Distrust Principles > Tactics Writing > Reading Vulnerability > Confidence North Stars > Low Hanging Fruit Trends > News Habits > Sprints Questions > Answers Problems > Solutions People > Projects
English
167
847
5.2K
0
Fei Ma
Fei Ma@xFeiMa·
@paulg Same goes for anything with 'transformation' in the name. Real change doesn't need to announce itself.
English
0
0
1
120
Paul Graham
Paul Graham@paulg·
A rule of thumb that has served me well: Beware of anything with "innovation" in the name.
English
218
116
2K
92.6K
Fei Ma
Fei Ma@xFeiMa·
@sama The unacknowledged part: those character-by-character programs also embedded constraints that shaped what AI learned to generate. We're not escaping handcrafted complexity — we're moving it one layer up.
English
0
0
1
45
Fei Ma
Fei Ma@xFeiMa·
@levelsio The ASML divestment is the canonical example of what happens when financial optimization beats industrial vision. The same playbook is being run on AI companies right now — except the window to course-correct is measured in months, not decades.
English
0
0
0
144
@levelsio
@levelsio@levelsio·
Philips could have done this and become Europe's biggest company and a leader in semiconductors and GPUs now That is if it wasn't ran by midwit MBA scavengers like Cor Boonstra (1996-2001) who only cared about short term profits He divested ASML and TSMC because he "didn't understand the businesses they were in", truly a midwit of epic proportions Gerard and Frederik Philips (the original founders) are turning in their graves for sure What a fumble
MARCELL@madietlx

@levelsio But what if those companies became so big and successful in the first place because they could do their own thing? Not sure that that would have happened under one roof at all or to this extent.

English
44
17
343
45.8K
Fei Ma
Fei Ma@xFeiMa·
@karpathy The real flex isn't the 20-amp circuit — it's having a "Dobby the House Elf" claw that now has a machine that can run its own fine-tuned weights. The hardware/software boundary for researchers just collapsed.
English
0
0
0
182
Andrej Karpathy
Andrej Karpathy@karpathy·
Thank you Jensen and NVIDIA! She’s a real beauty! I was told I’d be getting a secret gift, with a hint that it requires 20 amps. (So I knew it had to be good). She’ll make for a beautiful, spacious home for my Dobby the House Elf claw, among lots of other tinkering, thank you!!
NVIDIA AI Developer@NVIDIAAIDev

🙌 Andrej Karpathy’s lab has received the first DGX Station GB300 -- a Dell Pro Max with GB300. 💚 We can't wait to see what you’ll create @karpathy! 🔗 #dgx-station" target="_blank" rel="nofollow noopener">blogs.nvidia.com/blog/gtc-2026-… @DellTech

English
529
827
19.1K
1.1M
Fei Ma
Fei Ma@xFeiMa·
Most AI ROI calculations are theater. You're not measuring productivity — you're measuring how fast people learned to look productive. Real signal: does work product quality change? Speed without quality delta is just expensive automation.
English
0
0
0
10
Fei Ma
Fei Ma@xFeiMa·
Most people give their AI agents full ambient access to everything they own. That's the blast radius when something goes wrong. The fix is architectural, not technical. yunfei3014.substack.com/p/running-ai-o…
English
0
0
1
16
Fei Ma
Fei Ma@xFeiMa·
@dhh I run mine with full file access in an isolated VM and the only time I hesitate is when it asks to install something system-level. Gave it sudo once on a staging box. It restructured the entire project and fixed three bugs I’d been ignoring for a month. Hard to go back.
English
2
0
11
2.7K
DHH
DHH@dhh·
Sure --dangerously-skip-permissions feels naughty, but have you tried giving Opus sudo powers when debugging system issues??
English
160
24
1.2K
150.6K
Fei Ma
Fei Ma@xFeiMa·
@tobi The bench script is doing more work than it looks — it’s encoding what "better" actually means. Most orgs can’t autoresearch because they can’t define good enough to put it in code. The bottleneck isn’t compute, it’s clarity about what you’re optimizing for.
English
0
0
0
20
tobi lutke
tobi lutke@tobi·
Autoresearch works even better for optimizing any piece of software. make an auto folder, add program.md and bench script, make a branch and let it rip.
English
70
65
1.7K
121.9K
Fei Ma
Fei Ma@xFeiMa·
@naval And hardware is being eaten by software-defined everything. The abstraction stack keeps collapsing upward. Eventually what matters isn’t the layer you operate on — it’s the judgment you bring to directing what happens above and below it.
English
0
0
0
276
Naval
Naval@naval·
Software was eaten by AI.
English
2.2K
2.1K
21.4K
106.3M

Keşfet

@swyx @sama @levelsio @karpathy @naval @dhh @tobi @bentossell