Applied AI & research | models, agents & cybersec | still stacking sats
Won’t trap me
Joined November 2013
- Tweets51.5K
- Following1.3K
- Followers2.9K
- Likes222K
Bitduke retweeted
I was going to post some new sama quotes from this interview this morning, but prinz already posted a great summary. I will add that this is the third time I've seen Mr Altman talk about self replicating OpenAI datacenters. The first use case for OpenAI robots appears to be building more OpenAI robots and building more OpenAI datacenters. Mr A also wrote specifically about this on his blog a couple of years ago.
'Yeah, people talk a lot about RSI as AI that does AI research. There's very little talk about self-replicating data centers, this idea that you can use AI to control robots and run fabs and build turbines or solar panels and build data centers. But I think we have that within grasp.'
Sam Altman on automation of AI R&D, RSI, OpenAI's internal model:
1. "We don't know yet how [automated AI R&D] truly is going to work. We know today how to say: find me a 5% compute efficiency win."
(Note: a 5% compute efficiency gain that the model can find automatically is actually a big deal!)
2. "What I don't think anyone knows how to do yet is to say, go discover a new architecture that is as big of a single leap forward as the transformer was. And I think that's where things are heading."
3. "If you apply the same pipeline that we applied to get our models to be so good at math to AI research, it will probably work [for automated AI R&D]."
4. The bottleneck to getting good AI R&D results is shifting to asking the right questions and using things like swarms correctly, but answers to these questions will become much clearer in the next 3 months:
"One of the interesting things that people talk a lot about in OpenAI... is what are the questions that you're supposed to ask of GPT-7? Probably it's not enough to just say, make me RSI. And also, it's probably not enough to say, buy me a 5% compute efficiency win. I think this is one of the most interesting debates going on right now. And the answer is not just what question to ask it, but it's things like agent swarms. How are we going to use these models in a way that, like, actually maximizes what we can get out of them? I think we're going to learn a lot about that in the next 3 months."
5. OpenAI's extreme focus on model efficiency:
"I bet we could redo Navier-Stokes at 1/100 of the cost a year from when it was done."
6. More on OpenAI's internal model:
"There was one weekend when we got a new model... The name of the channel was "model-name-hold-my-beer". And people were just throwing that at the hardest problems they could. And people just didn't think the model was going to do these things. And we had this collective experience of people's hardest research problems from their academic career, the model was just knocking down one after the other, many different fields, it was incredible to watch... it was many of the problems we published, many other things."
7. The other loop is self-replicating data centers:
"Yeah, people talk a lot about RSI as AI that does AI research. There's very little talk about self-replicating data centers, this idea that you can use AI to control robots and run fabs and build turbines or solar panels and build data centers. But I think we have that within grasp."
Stan Druckenmiller just put his name on hyperliquid
but let’s go back in time first. Sept 16, 1992, he and soros shorted ~$10B of sterling. the Bank of England burned billions in reserves and hiked rates to 12%, then promised 15%. the market didn’t care, and by evening the uk had left the ERM. Soros’s quantum fund made $1B+ (!) and it went down as "the trade that broke the bank of england" so 34 years later, he’s backing trade[xyz] as ICE (the NYSE owner) brings 24/7 HL perps to the US
remember, earlier this year ICE was lobbying regulators against HL. now it’s investing in it and operating the venue (first they fight you, then you win)
the guy who broke a central bank is now betting on the market that never sleeps. yeah, CFTC still has to sign off, but big money already picked a side. crypto perps are eating every asset class
this one has me rethinking how ai reads
Cartesia, the startup behind voice agents and speech models, says its new model skipped the usual pre-cut word pieces, learned straight from raw letters(!), and matched smollm3-3b after about 1 trillion tokens of training while smollm3 needed about 11 trillion, so it's like, basically finished the book while the other guy was still on chapter one
it’s basically compression inside compression. and it even adapts mid-read, so a made-up word that keeps coming up gets split into fewer and fewer chunks
and if Cartesia is right that the tokenizer was quietly costing 10x the compute and data, that could mean a 10x smaller training bill for the same result
Bitduke retweeted
on security tasks, my gut feeling is still glm 5.3f > glm 5.3 > deepseek 4.1f 1017 > gpt 5.6 sol blue. folks are overhyping deepseek.
Bitduke retweeted
OpenRouter served 5.2T tokens yesterday. We did ~1.8T (1/3 of their volume) on just 64x B300 and 64x M890P. Where do all GPUs go? 😂😂😂
Bitduke retweeted
the 30 days of codex updates has felt like a huge larp unfortunately. but i believe in them for the next model lap!
Bitduke retweeted
Meet NVIDIA Nemotron 3.5 Transcribe 🎙️
An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint.
Watch it keep track of who said what when everyone’s talking at once.
1/ 🧵
was all set to sit down and play with Google’s new Carbon model this evening, then remembered it’s not even public yet. only a small group of cyber defenders gets it for now
bio/acc 2027?
this might be my last normal tweet before biotech goes vertical
i'm not a whistleblower. I don't work at Lilly, Novo or any of the big labs
but there is a massive shift happening in human biology right now.
it's sitting in press releases and almost nobody reads them
the last 6 months alone:
- a once-daily pill roughly doubled median survival in metastatic pancreatic cancer (13.2 vs 6.6 months)
- a one-time gene editing infusion cut hereditary angioedema attacks 87%, and 62% of patients stayed attack-free and off all other prevention for 6 months
- a one-time gene edit for cholesterol dropped LDL up to 62%, and it held for over a year
- a personalized mrna vaccine beat keytruda alone in a 1,137 patient melanoma phase 3
- retatrutide hit ~28% weight loss in a phase 3
- an ai-designed drug entered phase 3
- a pill that regrows hair passed its second phase 3
- an immunotherapy that could let some rectal cancer patients skip surgery, chemo and radiation just got fda approved
- a man lived 271 days off dialysis on a pig kidney
- human cells were caught passing chromosome-sized chunks of DNA to their neighbors, and it changes how the neighbors behave
we are on the verge of an era where the human body is a software problem and the update cycle gets shorter everyday
stick around, it's going to get weird
bio/acc
"crypto" just got hacked again, and this time it’s the data layer
@AlliumLabs is the company big players use to read the blockchain, turning raw onchain activity into clean data feeds. its own site says Visa, Coinbase, Metamask and Uniswap use it, plus 30+ of the fortune crypto 100. CEO Ethan chan says someone got unauthorized access to some of its systems on October 25 (interesting timing too, it came 4 days after two huge server flaws, i.e php2shell and node2shell)
core services are back, but customers are swapping api keys and uploaded data is still locked. Allium hasn’t said which systems were hit or whether any customer data was exposed
time for crypto to take cybersec way more seriously?
Bitduke retweeted
Team is working on this but it's not rolling out just yet, sorry for any confusion. Plan is to bring together best of both Chat / Work capabilities together at some point this year, but we're going to proceed carefully here and take our time.
We'll have more to share soon.
Google’s Cloud just doubled and the stock still got hit hard
Alphabet says Google Cloud revenue jumped to $30.7 billion, but it also spent $51.4 billion in one quarter on chips and data centers, so free cash flow came in at negative -$9.1 billion, shares fell around 10% after hours
and this looks like a GIANT bet to me. Google is eating the bill now, no doubt, and it’s still raising new debt and stock to pay for it, all because it thinks the ai payoff later could be enormous!
maybe it’s pain now and massive gains later (or maybe it’s the most expensive bet in tech, yolo)
what's next? a model powerful enough to listen to an entire crowd in a public square, transcribe every conversation, and identify who's saying what?
NVIDIA is releasing an open speech model that can transcribe a meeting, label who said what, translate it and answer questions about the recording:
Nemotron 3.5 Transcribe supports:
- Transcription in 40 languages
- Translation between English and 31 languages
- Up to eight speakers, with speaker labels in 13 languages
- Recordings up to an hour long
You can give it names and technical terms to guide the transcript, or ask questions about the audio. Audio Q&A is trained in English.
It has roughly 30B parameters, with about 4B active per token, and can run on your own NVIDIA GPU.
According to NVIDIA's launch materials, it ranks first among open-weight models on Hugging Face's English Open ASR leaderboard and leads VoiceArena's 14-system test of transcription with speaker attribution.
Built for recorded audio. Live streaming uses a separate model.
Super cool to see!
Model: huggingface.co/nvidia/NVIDIA…
Demo: huggingface.co/spaces/nvidia…
Blog: huggingface.co/blog/nvidia/n…
Bitduke retweeted
The thing is: the current change is SO fast that those who are waiting / hoping for directions on how to use AI efficiently for your work: it could come too late.
Team dynamics are also shifting very quickly: tenure no longer correlates doing better (faster / higher quality) work necessarily!
humanode got drained (through approvals)
@humanode_io says an attacker used a bug in its EVM on Oct 28 to take 12,773,243.03 eHMND from 436 accounts. nobody signed anything new, because the old token approvals were enough
so, what went wrong? Humanode’s custom precompiles, the built-in contracts that handle the native token, accepted delegatecall, and that let the attacker’s code act as someone else and spend their approvals. not rocket science, and tbh a bit of an embarrassing bug (this kind of bug has been widely known in crypto codebases for years)
the good news is Humanode says every owner will be paid back in full. it also says it will restore and rebalance the eHMND/USDC pool, with details coming through its official channels
the lesson? revoke approvals you don’t use. and yes, it’s zero days without a crypto hack again
Humanode incident update: the incident is contained, and all identified security issues have been fixed, including the EVM vulnerability and the ChainPort vault approval issue. The full investigation continues.
blog.humanode.io/humanode-in…
i don't really get the pessimism around. ai is opening up enormous possibilities for research, engineering, and innovation, i mean it's getting easier than ever to explore ideas and build things that were previously out of reach
there's never been a better time to be curious and ambitious.
Bitduke retweeted
It’s hard to see anything else happening but coding interviews going away.
This won’t mean that technical ability won’t be interviewed: but it interviews will be less predictable, more subjective and harder to prep for than the near-universally-disliked DSA interviews.
The coding interview is dead, at least for my team at Anthropic. I've seen enough to know that it no longer provides signal. I removed it from loops this month.
I have a spicy, optimistic take - I think there's an opportunity for junior engineers: felixrieseberg.com/hunger-be…
