Sabitlenmiş Tweet
domz
492 posts
domz
@DomZippilli
building businesses, usually with computers.
Virginia Katılım Eylül 2013
170 Takip Edilen152 Takipçiler
domz retweetledi
Attempted to write a Steam Engine hype at the era of Industrial Revolution as if it was the age of AI —
The steam engine breakthrough is insane right now.
Watt’s separate condenser + new GRPO optimization just dropped the 405 hp-class engine. We went from 7 hp → 70 hp → 405 hp+ in basically three years. One machine now does the work of 50+ men or water wheels — nonstop, rain or shine, anywhere.
Textile mills, ironworks, everything scaling 5-10x overnight. Productivity exploding.
This isn’t incremental. It’s automating physical labor at massive scale. Jobs shifting forever. Society about to look unrecognizable.
The Industrial Revolution isn’t coming. It’s here and accelerating faster than anyone predicted.
Terrified. Excited. Both.
What a time to be alive. 🚂💨
English
domz retweetledi
I'm so sick of reading em dashes and "it's not x, it's y."
I'm so sick of it, man.
English
Exactly! Maybe people got the wrong idea. It means that in eng, you focus on the things you can most immediately control; processes and technology. If the root cause is one person's technology contribution, we still say that; it's just that we look at processes for prevention (better review, avoid that pattern, etc.). Sometimes the issue really is performance, but that is a different conversation entirely.
English
a lot of this is blaming the tool instead of the engineering team, a lot of companies run a blameless culture - especially in ops - you don’t have to, you can tell engineers or your friends that their architecture is terrible and that their approach is dumb
a lot of engineering issues is a cultural issue
Mitchell Hashimoto@mitchellh
I strongly believe there are entire companies right now under heavy AI psychosis and its impossible to have rational conversations about it with them. I can't name any specific people because they include personal friends I deeply respect, but I worry about how this plays out. I lived through the great MTBF vs MTTR (mean-time-between-failure vs. mean-time-to-recovery) reckoning of infrastructure during the transition to cloud and cloud automation. All those arguments are rearing their ugly heads again but now its... the whole software development industry (maybe the whole world, really). It's frightening, because the psychosis folks operate under an almost absolute "MTTR is all you need" mentality: "its fine to ship bugs because the agents will fix them so quickly and at a scale humans can't do!" We learned in infrastructure that MTTR is great but you can't yeet resilient systems entirely. The main issue is I don't even know how to bring this up to people I know personally, because bringing this topic up leads to immediately dismissals like "no no, it has full test coverage" or "bug reports are going down" or something, which just don't paint the whole picture. We already learned this lesson once in infrastructure: you can automate yourself into a very resilient catastrophe machine. Systems can appear healthy by local metrics while globally becoming incomprehensible. Bug reports can go down while latent risk explodes. Test coverage can rise while semantic understanding falls. Changes happens so fast that nobody notices the underlying architecture decaying. I worry.
English
@cryptoanon69 @0xSero I'm able to run gemma4-31B over multiple GPUs using llama.cpp pipeline parallelism. 33 tok/s. If I had NVLink or a PCI switch, I am sure I could do tensor parallelism and go much faster. Anyway, just adding, multi-GPU dense can work if your speed isn't critical.
English
1. Dense Models - Slow and Smart
Example: Qwen3.6-27B / Gemma-4-31B
What it means:
- when a prompt is sent
- it gets tokenised (words are mapped to tokens)
- token generation starts
- the 27B means 27 billion parameters
- each of those parameters will be activated
- 27 billion matrix multiplications
- for every token generated
Active parameter counts are positively correlated with intelligence. That's why Gemma-4-31B is able to compete with Mixture of Experts (MoEs) 10 times their size.
2. Mixture of Expert models - Fast and Efficient
Example: Deepseek-V4-Flash / Qwen3.5-397B
What it means:
- when a prompt is sent it's tokenised
- it's sent to a router
- a router was trained to match prompts with experts
- experts are sub-networks of the model
- when found the experts are activated
- tokens are generated with only a fraction of the params
For example: Deepseek-v4-flash has 284 billion params 11x larger than the dense Qwen3.6-27b.
But only 13B of those 284B will activate per token, which is less than half of the size of Qwen3.6-27B
----
Dense Pros:
- Dense models are easier to train
- They tend to be smaller overall
- They can be very smart per token
Dense Cons:
- Competitive dense models are on average slower than their MoE peers.
- Less parameters to train and specialise.
MoE Pros:
- Can be much larger and be trained longer
- Faster token generation
MoE Cons:
- Larger vram requirements
- Harder to train
--------
Lmk if there's anything i'm wrong with or missing
English
domz retweetledi
Has there ever been a field that’s changed as quickly as programming in the last 18 months? The only thing more incomprehensible to me than that I used to hand write code is that there are thousands of programmers still doing it. An insane schism due to the speed of change.
English
I helped a friend experiment with something similar. Your intuition is what we found; benchmarks underperformed plain old typescript. All the "agent-first" design didn't overcome the novelty penalty. I think it was because the model has to make second order inferences to write the code.
English
Introducing Zero
The programming language for agents.
I wanted a systems language that was faster, smaller, and easier for agents to use and repair.
Explicit capabilities. JSON diagnostics. Typed safe fixes.
Made for agents on day zero.
English
domz retweetledi
There are two related, but distinct, problems with MTTR maximalism.
1. The distribution of recovery times could be heavy-tailed, and so the empirical mean could be far from the true mean.
2. Some failures are unrecoverable (e.g. durability loss).
Mitchell Hashimoto@mitchellh
I strongly believe there are entire companies right now under heavy AI psychosis and its impossible to have rational conversations about it with them. I can't name any specific people because they include personal friends I deeply respect, but I worry about how this plays out. I lived through the great MTBF vs MTTR (mean-time-between-failure vs. mean-time-to-recovery) reckoning of infrastructure during the transition to cloud and cloud automation. All those arguments are rearing their ugly heads again but now its... the whole software development industry (maybe the whole world, really). It's frightening, because the psychosis folks operate under an almost absolute "MTTR is all you need" mentality: "its fine to ship bugs because the agents will fix them so quickly and at a scale humans can't do!" We learned in infrastructure that MTTR is great but you can't yeet resilient systems entirely. The main issue is I don't even know how to bring this up to people I know personally, because bringing this topic up leads to immediately dismissals like "no no, it has full test coverage" or "bug reports are going down" or something, which just don't paint the whole picture. We already learned this lesson once in infrastructure: you can automate yourself into a very resilient catastrophe machine. Systems can appear healthy by local metrics while globally becoming incomprehensible. Bug reports can go down while latent risk explodes. Test coverage can rise while semantic understanding falls. Changes happens so fast that nobody notices the underlying architecture decaying. I worry.
English
A preview for Pro users: a new personal finance experience in ChatGPT.
Pro users in the U.S. can securely connect financial accounts, see where their money is going, and ask questions based on the information they choose to connect.
Your full financial picture, now in ChatGPT.
English
domz retweetledi
domz retweetledi
lets do more of this and less gambling apps. this is what startups should be all about.
Founders Inc@fdotinc
They reinvented the hearing aid by studying the human ear Normal hearing aid: $4700 Theirs: $20
English
@alexocheema @CrazySoft I'm new to this. Can you tell more about the clustering here? I have read some Exo blogs so maybe one has the answer. But it kind of sounds like the cluster does pipeline parallelism in this config?
English
@CrazySoft It really does work like that.
We’re moving tiny tensors ~10KB between the devices. I don’t know why you’re fixated on network bandwidth. The latency between the MacBooks with RDMA is ~8 microseconds.
English
4 x M5 Max MacBooks clustered with RDMA:
512GB @ 2456GB/s, $20k, 560W, quiet.
Find me a better deal, that I can buy today.
Ivan Kuleshov@Merocle
Cluster testing is currently in progress
English
@KhanAbbas201 Claude is great at stuff like this. Had it recover a wedged Linux box for me recently. This would be nerve-wracking though! "Claude, recover my 5BTC, make no mistakes and don't delete anything!"
GIF
English
This is literally insane
guy had 5 BTC locked in a wallet for 9 years
dumped his old college computer into Claude as a hail mary
Claude found the wallet file, debugged btcrecover's password logic, decrypted the keys, converted to WIF, recovered the funds
We are so unbelievably early
🍜@cprkrn
HOLY FUCKING SHIT OMG CLAUDE JUST CRACKED THIS SHIT, THANK YOU @AnthropicAI THANK YOU @DarioAmodei NAMING MY KID AFTER YOU 😍 blockchain.com/explorer/addre…
English
@QuinnyPig No jokes for the people at LinkedIn laid off tho. It feels awful y'all, I know. It does get better and you are gonna do great. Day at a time.
English
Doesn’t their product have a button you can’t get rid of to solve this precise problem?
Techmeme@Techmeme
LinkedIn says it has "implemented organizational changes"; a source says LinkedIn plans to cut about 5% of its 17,500 full-time workers (Reuters) (Visit Techmeme dot com for the link and full context!)
English
I think soon we will see a rotation of hosted inference opex into capex on local-agentic-capable laptops ($AAPL leading there), and moderately sized self-hosted inference stacks for workloads where the token count is high and you don't need frontier capability. Gemma 4 is great.
Laura Bratton@LauraBratton5
New: @ServiceNow is the latest major public company to say it’s blown through its full year budget for AI coding tools from Anthropic in the first few months of 2026, just like @Uber CTO @praveenTweets said abt his company. “It’s a really hard problem,” CIO Kellie Romack said.
English
@ishverduzco @akothari If only we could slap a turbine on so-over/so-back cyclic energy.
English
One interesting trend over the past year is how quickly vibes complete the full circle. In just the last few months, we’ve done multiple round trips:
- PMs went from “the role is dead” to “we still need PMs” to “PMs might be the most important function.”
- Software engineering went from “AI will wipe out coding jobs” to “new grads can’t get hired” to “software hiring is booming again.”
- SaaS itself went from “software is dying” to SaaSpocalypse to “actually, AI is software too. We are so back!”
English
I think the intelligence in LLMs is more familiar than not, and prototypes work great with them. So does iteration, memorizing learnings, and NOT doing waterfall.
Matt Pocock@mattpocockuk
The more I replace plans with prototypes, the better the outputs Who'd have thought that low fidelity prototypes were better than walls of spec Oh yeah, the entire industry for 20 years Stop going against decades of knowledge because someone in SF shipped it as a 'mode'
English