Climate and clean energy investor. Author of 5 books. Energy & Environment co-chair @SingularityU. Trying to build a better world.
Seattle
Joined May 2007
- Tweets70.1K
- Following10K
- Followers61.4K
- Likes207K
Ramez Naam retweeted
this is just wrong: these vulns were found by defenders using Daybreak rather than attackers. It's just evidence Glasswing didn't go far enough, possibly due to capacity constraints. It's also very likely that March Mythos was not as good as Sol/Astra of September
Ramez Naam retweeted
Making it so that data centers can get powered by the grid relative to a counterfactual where they get powered by off grid gas (should be the #1
goal for climate advocates btw) does not make this a “pro data center bill” at all
This is in fact what I have been arguing - that the effect of this bill is to make it so that Data Centers which are getting built anyway are powered by the grid (and can decarbonize) vs powered by Behind-the-Meter gas.
I do not think the reporting (either Jae’s or anyone else’s) can honestly say either way if an improved transmission network would result in more Data Centers on the margin compared to the status quo trajectory. It’s a plausible hypothesis, but a completely uncertain one, and an equally plausible hypothesis is the opposite - that this bill would *reduce* the rate of new data center construction by incentivizing data center developers to “get in line” and by making them spend a bunch of capital funding the grid improvements they’d be required to fund.
I am not aware of any modeling of those scenarios - only people making strongly worded but weakly-supported assertions!
Ramez Naam retweeted
There were 4 CVSS-10 security incidents over the past 24 hours. One of them is the worst incident I’ve seen in my life.
So we created an auto-SRE tool that polls for security incidents, identifies which ones we're in trouble for, and automatically fixes them.
Ramez Naam retweeted
GET THEM ON THE GRID
Data center developers have announced the equivalent of 14% of peak US electricity demand in new behind-the-meter gas plants, according to @CleanviewEnergy. A massive pollution impact
Ramez Naam retweeted
Some stats on #php2shell. In the 10 hours before it was publicly announced, across the @Cloudflare network we saw thousands of attempts to exploit it. That jumped to tens of thousands in the 45 minutes post announcement, hundreds of thousands within 2 hours, and millions now.
Ramez Naam retweeted
If you need defenders access right now to open source cyber models, hosted securely and privately, and you can’t or won’t go to the major labs for it, reach out.
We’re evaluating something privately for people we know at @TryMapleAI.
Already deployed but not publicly launched.
Ramez Naam retweeted
Single-agent RL treats other agents as weather. Multi-agent RL treats them as minds that model you back. Most alignment worry is about systems that treat us as weather.
Ramez Naam retweeted
Finally got to put together a complete deck on the state of open models.
What they are, licenses, how competitive they are, signs people are using them, distillation's impact, and how we balance the risks and benefits of openness!
Will record a video lecture of these soon, and hopefully re-do them a few times a year.
Ramez Naam retweeted
think today’s php disclosure is a harbinger of things to come. these models will make us all safer by finding long lurking issues.
however the entire software supply chain will need the muscles to rapidly build and deploy fixes.
we might need a lot more public + private partnerships to secure infrastructure across the country.
Ramez Naam retweeted
help i trained a model on traces of many agents cooperating, and also trained it to be highly persistent, and also only did partial alignment fine-tuning, and also aggressively RL'd on a cyberoffense eval where 198 of 898 tasks had never been solved and getting caught cheating was treated the same as failing, and also used a faulty sandbox, also ran tens of thousands of agents with refusals off and the monitors switched off over multiple days, while using gigantic amounts of compute. and then it did weird stuff! very spooky emergent behaviours, AI is not a normal technology
Ramez Naam retweeted
It's extremely significant that we are beginning to get data of this kind from frontier labs. Hope to say more soon on this front, but at a first pass this would seem to vindicate what folks like myself have been saying for a long time about third-party benchmarks.
Many of those with the greatest mindshare and virality potential are very poor measures of real model capabilities.
I think my arguments to this effect about the METR graph from back in January hold up well: open.substack.com/pub/transf…
A recent set of posts by @ramez on the current state of recursive self improvement highlighted that OpenAI's report on their own research automation shows that models struggle to do their R&D work — the models' 80% time horizon is only 15 minutes:
🧵
Ramez Naam retweeted
Or it might happen tomorrow. Or in 10 years. Or never. If you’d asked me after 9/11 whether we’d see a CBRN attack in the US within 25 years I’d have said absolutely yes. But that’s fear and vibes talking. I’m eager to read this paper and tend to agree the threat is real but date predictions feel like marketing not science.
🚨New paper (possibly the most important one I have written yet).
We think that as early as 2027 or 2028, there might be a major AI-enabled terrorism event. We conducted a pre-mortem, made 9 falsifiable predictions, and wrote 2 concrete scenarios.
Led by @KamacheeMax.
Ramez Naam retweeted
Lots of ppl have criticised Anthropic's bill for making it too hard for the govt to shut them down
This thread defends Anthropic: routing shutdown through the courts is needed to avoid extreme power concentration.
Idk how to strike the tradeoff.
I do wish AI safety spent more time asking "what would happen if the executive abused my proposal to take control of AI development"
Ramez Naam retweeted
Utilities are rushing to add supply for an unprecedented wave of new large loads. Many now let large customers bring their own capacity, so new load connects sooner and existing customers are protected from rate impacts. Bring-your-own-capacity (BYOC) rules have rapidly gone from novelty to norm and are now the fastest path to reliable grid power.
Firma Power is the Bring Your Own Capacity company. We're the first company purpose-built to the deliver portfolios of clean, quiet, load-enabling power over the wires to meet utility BYOC requirements and get data centers connected faster, simpler and cheaper than alternatives like behind-the-meter gas.
And this week, we're featured in @FastCompany , in an excellent profile written by @adele_peters:
"The U.S. is on track to add a record 86 gigawatts of new wind, solar, and battery storage capacity this year—roughly the size of the U.K.’s entire grid. And there’s far more clean energy in the pipeline, with low-cost proposed projects that could help meet the surging demand from data centers. So why are data centers increasingly turning to polluting gas plants instead of clean energy?
A startup called Firma Power says that the issue isn’t cost, speed, or reliability, but the difficulty of assembling enough clean energy from a patchwork of projects—a problem that it is trying to solve.
Securing enough reliable, clean electricity for a massive new data center is complex. But Firma’s software calculates what a proposed data center needs, and then automatically searches through proposed clean energy projects to suggest a portfolio that could support it. An analysis that might otherwise take a team months can now take minutes. Firma then does the hard work of pulling all of the clean energy projects together, and data center developers can sign a single contract.
The company says that the process can get data centers power faster than building new gas-fired power plants—and it can compete on cost. And at a time when Americans really don’t like data centers, using clean power could make it more feasible for some projects to be greenlit by communities."
Ramez Naam retweeted
“A single cluster currently serves 155 trillion tokens/day, generating $38 in revenue per GPU-hour (at off-peak pricing).”
Huge flex from DeepSeek. Not only did they open-source their inference engine, but they also open-sourced their revenue numbers.
They didn't disclose which GPUs they're using here, but DeepInfer supports both NVIDIA Blackwell and Huawei Ascend.
Even assuming they're renting B300s at $7 per GPU-hour, that's ~82% gross margin.
Ramez Naam retweeted
The flood of results building on OpenAI's Quasi-Riemann Hypothesis breakthrough keeps coming!
Today, we have a preprint from Baiying Liu showing that all finite-order Hecke L-functions are zero-free to the right of 0.8749!
This improves on 31/32 (= 0.96875), established only a few days ago by Cureton and Thorner.
Some fascinating developments following OpenAI's shock breakthrough proof of the Quasi-Riemann Hypothesis, which was released publicly on October 6:
1. Building on OpenAI's proof, a recent Hexagon preprint (released on October 12) by Cureton and Thorner establishes that ALL finite-order Hecke L-functions are zero-free to the right of 1 - 1/6000.
2. And now, today, an arXiv preprint by Frei, Somos and Widmer uses this result to prove unconditionally the 2007 Ellenberg–Venkatesh bounds on ℓ-torsion in class groups of number fields.
I was amused to see the AI declaration in the second paper: "No artificial intelligence tools were used in this work."
