Damir Marusic (@dmarusic) header
Damir Marusic (@dmarusic) profile picture

Assignment Editor, @PostOpinions. Co-founder of @WCrowdsLive. My opinions are my own, and should not discourage you from pitching me.

The District
Joined March 2007
Photo by Damir Marusic (@dmarusic)Photo by Damir Marusic (@dmarusic)Photo by Damir Marusic (@dmarusic)Photo by Damir Marusic (@dmarusic)Photo by Damir Marusic (@dmarusic)
Damir Marusic retweeted
Just posted: My piece in the Washington Post about why we need a national AI pause. washingtonpost.com/opinions/…
1
7
1
20
582
That this is written earnestly tells you everything.
A dear friend told me last night that AI's returns won't come fast enough to justify the spend. They were thinking about enterprise adoption and on that, they're right. Businesses adopt slowly: compliance, regulation, resistance and, well, people. If you consider the gap between what's being spent on infrastructure and what's coming back in revenue, it's a fair question, but it's the wrong one. *Enterprise adoption is not the bet.* The labs are not selling software! They are racing to build SuperIntelligence. An 'Infinity Machine': a system that takes humanity's slow, unscalable, deeply inefficient process of discovery and accelerates it without limit. So the ‘prize’ is not adoption by the Fortune 500 or billions of consumers — although that inevitably follows. The prize is the ability to 'solve and understand everything.' We have already seen the opening act. In coding, software, cyber and maths, AI's gone from 'assistant' to 'author' to 'creator' in a matter of months. It is not helping in those fields, it’s redefining them. In the wise words of @ilyasut, ‘if you value intelligence above all other human qualities, you’re going to have a hard time.’ Coders and mathematicians are already reconciling with this reality. And while the ‘average human’ might still not even use AI, this reckoning is coming first for those who work at frontier of human knowledge, where our understanding is still so small: in physics, engineering, biology and computing itself. The people building these systems now say SuperIntelligence across *every domain* could arrive in the next 12 months. Yes: by the end of 2027. Even if it takes five years, nothing about this logic changes. (Sentience is a separate question. More on that soon.) BTW, the labs will also stop prioritising enterprise Intelligence, because the compute crunch is coming. And when it does, the labs will hoard it for their internal research. (IE, to build ASI/AGI/SuperIntelligence.) As capital tightens, politics hardens and public resistance grows, the labs will sell less and less Intelligence. Every token they sell is a token not building their own Infinity Machine. And if one lab gets there first, it gets first use of SuperIntelligence. First pick of what it discovers across all of science and engineering. And the power to decide what gets built, what gets shared, and on whose terms. The ROI is understanding more of everything, everywhere, all at once, and then for infinity. That’s the ROI.
1
4
1,086
Damir Marusic retweeted
OpenAI has released data on AI research acceleration that's far more detailed, granular, and longitudinal than Anthropic. And in this data we don't see anything like runaway super-intelligence. I respect the METR team who went into Anthropic quite a bit. But without this level of data transparency, it's hard to be convinced that Anthropic really is seeing an AI-driven acceleration. To be clear, the METR team had their hands tied by Anthropic, who redacted the clearest evidence. The data Anthropic has released and allowed METR to release is insufficient for independent conclusions. Anthropic says that they expect to hit the early stages of RSI (and pass one of their safety thresholds) in a forthcoming model. Given such a statement, the responsible thing for Anthropic to do is to release data that matches OpenAI's in granularity and over a similar span of time, so anyone can independently analyze it to understand the capabilities of Anthropic internal models and their trajectory.
Seems like the OpenAI research acceleration blogpost is not what you would expect if we were definitely on the cusp of artificial superintelligence - models that are more cognitively capable than ~ALL humans on ~ALL tasks. Agent workdays have grown very jaggedly, swarm sizes haven't grown very much, and spending on tokens hasn't actually outpaced spending on human workers* although the growth rate is clear. There also hasn't been an acceleration in research experiments per experimenter, or in the internal ECI scores of OpenAI's models. Perhaps most importantly (and most surprisingly), the task success rate for <15 min tasks hasn't grown very much at all, and the share of tasks done by agents across various time horizons is both low and not growing for the most part. It's possible that Anthropic is *much* further along in automating AI R&D than OpenAI, and that METR saw this in their investigation. It's also possible that we see nothing much right up until a shocking acceleration in capabilities, like we saw in mathematics. We certainly can't rule out the possibility of an intelligence explosion. But there's been a huge belief cascade in AI in general and AI safety in particular that seems largely unwarranted given the evidence that we have today. It's entirely possible that we can't fully automate all AI R&D tasks in the imminent future, or that full automation of AI R&D doesn't massively accelerate AI capabilities progress, or that capability gains concentrate in easily verifiable tasks and fail to generalize to the messy real world. These are all open research questions
10
33
2
238
18,033
Damir Marusic retweeted
I’m something of a student of the Bond market myself
2
1
28
744
Damir Marusic retweeted
Dunkin update: @dmarusic has opted for an environmentally conscious cookie butter latte. I’ve gone for ol’ reliable: blueberry cold brew with cold foam. Bracing for apocalypse with a “Doom tin” for our colleagues filled with the only doomsday rations I need: 25 munchkins.
2
1
1
15
1,690
Damir Marusic retweeted
AI solves disease and aging, and everyone starves after having eaten everything, including the grass and cockroaches. p(doom)?
1
3
548
The main problem with universal empathy is that while it doesn't exist, it's incredibly alluring to pretend that it does.
My guess is that the centrist focus on climate change in the 2010s was basically just normal group psychology. A thing to worry about to sound serious and adult. Doubt many were actually personally that anxious.
2
518
Damir Marusic retweeted
My guess is that the centrist focus on climate change in the 2010s was basically just normal group psychology. A thing to worry about to sound serious and adult. Doubt many were actually personally that anxious.
I'm baffled about the centrist position on climate change at this point. Is the idea that it isn't happening anymore? Or that we don't need to take drastic measures to prevent it, even if they are unpopular?
57
25
12
572
84,366
Damir Marusic retweeted
Another common misconception about a dirty bomb is that the “dirty” aspect makes them far more dangerous than normal bombs. The blowing up does damage. The radiological dispersal is almost trivial. The concept itself is designed to terrorize. Don’t fall for it.
This is a common misconception about a dirty bomb. It’s not a nuclear weapon. You don’t need uranium, let alone highly enriched uranium. It requires radioactive materials that are often used in medical, agricultural, industrial, and other applications. Not easy to acquire, but certainly easier than uranium.
2
2
482
Damir Marusic retweeted
Some quick reactions to the recent Dwarkesh podcast with Paul Christiano. Tl,dr: bullish on rapid capability gains, more skeptical of the leap to takeover, and wary of emergency overreach.
29
61
16
510
37,240
Damir Marusic retweeted
You will notice that Gemini Carbon thinks differently than Claude and GPT. It often comes up with entirely different but more clever solutions than competitor models. This is precisely because we never distilled other models. Distilling other models creates a poor imitation rather than unique intelligence.
Google has proven that when you’re months or even a year behind, the solution isn’t to distill models. and they proved that good research takes time and I have high hopes for SSI as well :)
326
473
115
11,090
533,573
I'd think that training a local LLM to write the best heds should be easy. Actually, that's a fun weekend project...
SEO and AI have murdered the art of headline writing
1
461
AI solves disease and aging, and everyone starves after having eaten everything, including the grass and cockroaches. p(doom)?
1
3
548
BREAKING: Jensen Huang is in the datacenters ripping the Nvidia cards out - causing worldwide outages - saying “they’re mine anyway”. He removed his leather jacket, revealing he’s all greased up, and they can’t get a hold of him because of the grease
192
4,797
93
64,925
945,242
Damir Marusic retweeted
I keep saying that the time to ring the bell on the Singularity will be when all the AI nerds start sporting thick, lush heads of hair
Can these labs start by just solving one disease first and then maybe we can believe them to solve all the rest
1
5
37
1,872
When @ChristineEmba and I recorded an episode this weekend, I was worried that the apocalypse theme would be played out by Tuesday. I guess I was wrong! wisdomofcrowds.live/p/dont-w…
1
6
441
Most plausible theory yet. (And ffs follow @jon_stokes if you're not yet doing it for whatever reason.)
I have an alternate theory on the current disruptions that I'll float: There really is a spike in updates/upgrades breaking things, & this is most of what's going wrong right now. Some backstory:
1
2
11
5,150
Epic thread, and really perceptive about something else that I've been thinking a lot about: one of the core premises of the AI world is that we're ushering in the end of human politics by creating a hyper-rational machine god. Not how it's gonna play out. History is conflict.
I like @ciphergoth. He's an honest, curious guy who I disagree with on many things. And I like @politicalmath. But I'll be blunt: let's all cut the "smol bean" act. The "we're all collegial, quirky TPOT nerds on a forum" era is over. Play for civilizational stakes or go home.
1
1
15
1,225
Damir Marusic retweeted
based on what I’ve seen this last week it really seems like there’s been a huge increase in recency bias
28
200
8
2,320
54,751
So probably some massive vulnerability was discovered (but somehow not publicized) and now everyone is patching? Maybe that’s it.
1
1
13
1,507