@joefioti

ceo @luminal_ai delivering superintelligence to the proletariat.

 San Francisco, CA
luminal.com
Joined June 2015
  • Tweets2K
  • Following517
  • Followers3.5K
234Media
your sanity is my moat.
4
1
49
19,346
Growing up is realizing anyone with this opinion isn’t handling real traffic.
sadkat
@sadkatwtOct 21
17
726
Can someone from Nvidia explain why enqueuing kernel launches to a stream has more overhead than a cuda graph? Fundamentally this shouldn’t be the case if my cpu can run ahead of my gpu, right?
4
2
1
67
8,499
Swing by the lowest budget booth at PyTorch Conf tomorrow
2
1
47
1,270
“Claude put the Luminal dashboard in my living room”
1
17
831
I am starting a robotics company in my spare time called Big Gripper™️ that takes cheap servos, webcams, metal, plastic, hooked directly up to Luminal Cloud, and lap all the custom robotic model companies out there. Might point codex at stripe atlas and let it rip.
Jim Fan
@DrJimFanOct 16
We are open-sourcing V1.0 of our agentic robotics framework, ASPIRE, with full real robot skill discovery and self improvement capabilities! On the day Astra arrived, our lab switched the backend model and saturated all the benchmarks to ~99% over night. And the frontier is already on the move: - Opus 5.5 slightly edges out Astra on our robot simulation bench. - Sonnet 5.5 hits the Pareto frontier of cost vs. robot capability, $12/task. We first introduced ASPIRE and /robot skill library in June, way before the big models could reliably do robotics. A rising frontier tide lifts all boats, but Astra was a 0->1 tsunami on many tasks. Now you can also feel the Physical Bitter Lesson at home like we did on that fateful day ;) links in thread
4
1
19
1,456
Is there a real moat for all of these sandbox companies? Feels like everyone is pivoting and the product is always the same??? - Fast boot time - High clock speed - Scale up and to zero But like beyond lightweight microVMs it feels like the technical depth is fairly shallow? Like everyone is running on the same CPUs, and you can't really optimize the actual workloads because its just random agent-generated code at runtime. Aside from race-to-the-bottom on pricing, how do you compete?
19
3
2
88
9,331
growing szn. need compute.
Jake Stevens
@stake_jevensOct 16
We're growing again! Doing over 50 trillion tokens a month @luminal_ai. If you can get me access to B300s in the next week or two, you'll be my best friend
1
1
1
24
1,719
rewriting reality with lots of flops...
Luminal
@luminal_aiOct 15
Krea 3 is now available on Luminal Cloud! Get access to some of the best image editing we've seen yet! Just look at how @krea_ai brought the drawing below to life: Check it out on our public cloud 👇
4
446
We've been growing insanely fast on private dedicated deployments and are just now starting to open up our cloud to the public. ~2.3 trillion tokens per day. Not happy till Luminal-compiled inference drives a few quadrillion tokens per day.
Brett Harrison
@BrettHarrisonOct 14
The Liquid Inference router network is growing quickly. We now have 28 verified companies serving >1000 models, offering capacity dynamically and updating quotes in real-time. Average time for providers to onboard and integrate models is 3 hours, full verification within a day.
3
5
66
4,640
Might be a moment of hubris for me but I’ll still say it: skill issue. Noam and co had a really innovative model arch, and it’s wild to me they couldn’t figure out how to spend less on compute than they were charging users. Modern batch inference is insanely efficient. Just raise prices and use smaller models.
Jason Koebler@jason_koeblerOct 13
I pulled the Character bankruptcy docs. They show: - They were spending $14 million/month on compute alone at a time their revenue was around ~$1 million/month - Currently losing $9 million a month - They admitted they enshittified & that it didnt work 404media.co/character-ais-ba…

Character.ai's Bankruptcy Documents Show the Economics of AI Don't Work

Character.ai enshittified its product but was still burning tons of cash, its bankruptcy documents say.

404media.co
4
2
34
4,720
I keep talking about this. High-batch inference is already are nearly as efficient as the human brain, and we’re on track for at least another 10-100x efficiency gain.
David
@DavidSHolzOct 11
Let's say a frontier model is a 10T MoE on GB300 NVL72. That's ~5 Joules/token or ~50 W at human thinking speed vs our brain at 20 W. The energy grid is $0.08/kWh and eating rice $0.50/kWh. So on an energy-per-thought standpoint rn we're surprisingly similar to frontier models!
1
1
1
71
3,871
luminal is serving ~45 trillion tokens per month, and 5 million images per month. we run more than an exaflop of compute. still feels insane to me.
Matthew Gunton
@matthewjguntonOct 9
we are now serving roughly 1.55B tokens a minute
5
4
1
77
4,416
Who is making The Simple Robot Company™️? - Three or four way opposable grippers - Wheels - Shitty cameras - Shitty (imprecise) servos - Piped directly into GPT-6 Astra It would be so straightforward and you can lap 90% of existing robot companies.
14
839
We’re looking for compiler / kernel engineers who want to work on hard performance problems with a small team in SF. Come build with us.
Luminal
@luminal_aiOct 7
Replying to @luminal_ai
Cloud Inference Engineer - luminal.com/careers?job_id=8… Compiler Engineer - luminal.com/careers?job_id=8… Senior Compiler Engineer - luminal.com/careers?job_id=9…

Luminal - Inference at the Speed of Light

Luminal is an AI inference compiler that compiles and optimizes AI models for GPUs and ASICs, delivering the fastest, highest throughput inference in the world.

luminal.com
7
13
135
6,570
Our first launch video cost nothing and took us about 5 hours. Filmed on our phones in our apartment. The only thing we paid for was a mic because people hate bad audio.
2
3
24
2,233
Being able to say “I don’t know” for things you don’t know on an interview is a seemingly low bar that many fail to clear. They’ll twist themselves into pretzels coming up with an obviously wrong answer rather than just admit what you don’t know.
5
1
55
2,117
Joe Fioti retweeted
Jake Stevens
@stake_jevensOct 1
There’s always a higher dimensional move to be made when you feel stuck
David
@DavidSHolzOct 1
when i was a teen i was a little emo, and i think it was because i felt like i didnt control anything important about my life. looking back with knowledge now, there were a nearly infinite number of higher agency paths i could have taken. my life was always more free than i knew
4
9
2,135
Joe Fioti retweeted
Avik Chaudhuri
@__avikSep 30
Had a great conversation today with a colleague on whether compilers are dead. It left me with more optimism than I had going in. Basically the argument is this. Given a program, an agent can probably optimize the shit out of it burning a bunch of tokens. But what happens when you have multiple coordinating agents? Maybe they will discover the power of abstraction. Certainly they need to divide up work and collect it back. They would need to trust each others' work. Then we'll get greedy and ask them to optimize many programs. They can continue burning tokens on each program, but soon they'll discover they're doing the same thing over and over again. So they'll discover compilers. Then they will realize that they need to re-prove at a lower level what seemed correct at a higher level. So they will discover correct-by-construction compilers. The future will be fine. Languages may change, but the concepts will endure. The same things that humans need to collaborate will be needed when agents collaborate.
4
7
1
39
2,422
Joe Fioti retweeted
Luminal
@luminal_aiSep 30
Ideogram 4.5 is now available on Luminal. 4.5 is the first precise image editing model that truly outcompetes head to head with Nano Banana Try it out on our public cloud today ⬇️
3
6
1
19
1,676
At Intel, we spent millions throwing hardware fixes at bad software. @luminal_ai exists to make the compiler take on more complexity so hardware does less.
8
8
2
152
10,245
Load more