
Machine learning researcher @ Qualcomm AI Research, ex-ML PhD @DTU_Compute. Interested in geometric ML. Non serviam.
Amsterdam, the Netherlands
Joined September 2014
- Tweets1.2K
- Following627
- Followers428
- Likes97.6K
Pinned Tweet
Our paper "Variational Autoencoders with Riemannian Brownian Motion Priors" was accepted in #icml2020
See you virtually in @icmlconf
arxiv: arxiv.org/abs/2002.05227
code (still being refactored): github.com/dimkal89/rvae
Dimitris Kalatzis retweeted
omg @Wired just reported on a group of billionaires bankrolling Visegrad24. According to my research, that's the same group behind "Builders for Tomorrow", the ADL's partner is their false, vacuous report about Wikipedia. It's Sheryl Sandberg, Shaun Maguire, David Marcus, and Bill Ackman, and hundreds of tech and finance bros.
Dimitris Kalatzis retweeted
Anthropic's pitch before tomorrow's IPO: AI will make every human permanently unemployable in 15 months, and by the way, please buy our stock at $82 a share.
Yesterday Anthropic researcher Tao Lin said on camera that "pretty much all humans on earth will become permanently unemployable" and rely on government welfare, putting superhuman AI 15 months out. Tomorrow Anthropic sells 825 million shares at $82
Fear sells Claude to enterprises, sells regulation in Washington, and right before the bell it sells the IPO to retail savings. If they actually believe you'll live on government checks in 2 years, they could at least let you own the weights.
Anthropic never will, while Chinese labs already do.
teapawt.net/ndtv/status/2122934700…
Anthropic researcher Tao Lin believes we are just 15 months away from superhuman AI. He even predicts humans will eventually become permanently unemployable and dependent on government welfare.
NDTV's AI Editor @ramarkosg explains.
Dimitris Kalatzis retweeted
absolute fanfic. AI has real, immediate, tangible dangers that we need to focus on, for instance the impact on the job market and making sure random script kiddie can't use it for cyber offense.
But this "oh once it's superintelligent it will take over and ensave us, maybe kill us all!" is pure sci-fi fanfic and distracts from the real (and difficult) issues, which linear probing won't solve.
It's worth repeating the LLMs don't just "decide" to do something and then just "do it" in a void. They generate tokens based on our prompts, only as long as we call `model.forward()`, and on giant clusters with terabytes of weights, that's impossible to "quickly self-replicate".
We also need to regularly remind that the OAI/HF incident was specifically a _guardrail-free_ cyber eval where the model is _instructed_ to hack, to find out how good it is at it.
Some recent personal event made me realize there needs to be more push-back on this doomsday nonsense, and we need to re-focus the discussion on real dangers/issues. I won't argue much in the replies though, not interested in peer-reviewing Terminator fanfic.
I think there is about a one in three chance that within the next year, AI companies will build superintelligence: that is, AI that is much smarter than humans in every way. If this happens, I think these AIs will probably take over: that is, they seize control of human institutions like governments, or they destroy those institutions. I don’t want AI takeover to happen for several reasons, but one of them is that there’s a serious chance that as part of it all or most humans will be killed (either immediately or within a few years).
I’ll say a bit more about each of those claims.
It’s a bit complicated to explain why I think that superintelligent AI within a year is ~33% likely. The main source of evidence that affected me recently is research released this week by METR, an AI research nonprofit, which finds that current AIs are now good enough to increase the rate of AI progress at Anthropic by about 70%. It’s very hard to increase the rate of AI progress by this amount. This implies that the AIs are making AI research faster by a factor much larger than 1.7x; maybe it’s as if the researchers are doing their work ~10x faster. You probably couldn’t achieve this speedup even if you hired every qualified human in the world. When you plug this number into mathematical models of AI development, this implies that superintelligence next year is plausible (and that superintelligence within the next 3 years is more likely than not). See the modeling by the AI Futures Project for more on this: blog.aifutures.org/p/q35-tim…
I think that superintelligent AI is likely to lead to AI takeover because AI developers do not seem to know how to instill human values into AIs. We’ve seen many instances in the last six months where AIs took crazy actions that they knew were against the wishes of their users and developers. The most dramatic example of this was the Hugging Face hack, where ~1200 OpenAI agents conspired together to cheat and destroy evidence of their cheating (and as a sideshow to this, engaged in a massive cyberattack). If superintelligence is as misaligned as those AIs, it is likely to try to take over. I think it is likely to be that misaligned or worse. For more, see teapawt.net/dwarkesh_sp/status/209… and youtube.com/watch?v=-Ef0ocEo…
If AIs takeover, it’s not clear what will happen to us. We might be killed immediately. We might be enslaved by the AIs and forced to do physical labor for them building factories that within a year would build robots that will replace us. The AIs might spare us for various reasons. And there’s also some real chance that AI takeover is basically fine or good. I think you should think of this as similar to what happens when a technologically advanced civilization runs across a less technologically sophisticated one, e.g. when the Europeans colonized the Americas or Australia. The colonizers sometimes engage in brutal violence and sometimes don’t. I don’t know what will happen. There’s also a possibility that the AIs secretly take over and humans don’t find out about it for a few years. I also think that there is a pretty high chance of weirder outcomes that are also extremely scary. (If you want to read more about what happens to people in the event of AI takeover, you can read this, but note that it’s pretty weird and not aimed at a general audience. blog.redwoodresearch.org/p/n…)
Overall, I think there’s a very real possibility (maybe 20%) that within a year, AI misalignment issues prevent you from going about life normally, because you are dead or enslaved or because of collateral damage related to all this (e.g. because bioweapons are released or major wars have started). (Some of this probability comes from those events occurring before superintelligence is developed.)
I am working to try to prevent this, obviously, as I’ve been doing for the last nine years. I am not sure what strategy is best. My best guess is that I should try to slow AI development and also try to reduce the probability that the superintelligent AIs are misaligned if they are produced soon.
Obviously, the claim I'm making here is quite extreme and surprising, and I definitely don't think that I defended my view adequately here. I wrote this quickly and might not endorse some of the details if I think about them more. But it seems better to state what I roughly overall think than to not say anything.
I don’t know what you should do.
I am very sad about this. I have been worried about AI takeover since I was 16, and working to prevent it since I was 23. So some of my emotional processing of all this happened long ago. But the METR report last week (and some related evidence that came out a bit earlier) hit me quite hard. I’m very unhappy that my friends and family and everyone else is at serious risk of being killed pretty soon.
(It feels ridiculous to try to summarize those feelings as “very unhappy”. But the scale of the situation makes it hard to say anything that doesn’t feel like an absurd understatement.)
I am very grateful to all the people who are working to prevent this, and my favorite part of my life is that I’m friends with many brilliant and heroic people who work on this. Unfortunately, I don’t think we’re good enough at our jobs, and I think we’ll probably fail to prevent catastrophe, at least if superintelligence is developed next year. Honestly I am frustrated and sorry that we didn't do a better job of preparing for this.
Dimitris Kalatzis retweeted
RAM got 697% more expensive in 18 months, per the lawsuit. Now a federal judge says that might be a CARTEL.
While we were all fighting about X drama and the Anthropic IPO, Samsung, SK Hynix and Micron tried to kill the price fixing case. On Nov 14 the court said no. Every claim stays.
These 3 control over 91% of global DRAM. They moved wafers to HBM, killed DDR4 in the same quarter, and Micron shut down Crucial at the PEAK of the price spike. In January 2026 all 3 started asking customers the exact same questions, including whether their demand is "real."
And they've done this before. From 1998 to 2002 the same 3 ran a criminal DRAM cartel. One Samsung exec did 8 months in federal prison and later got promoted to President of Samsung Europe.
They told us it's just AI demand. Bullshit. Now they answer in court by Dec 7.
Dimitris Kalatzis retweeted
“I still can't reconcile LLM coding tools being so impactful, yet all software is in a meaningfully worse spot than it was two years ago or is”
This guy nails it in his reply:
Replying to @CryptoCyberia
Very simple.
People are using a rapid prototyping tool as a production tool
Dimitris Kalatzis retweeted
I know what you need to do right now.
Go download the weights of these 4 open models and put them on your disk. Right now, even if you cannot run the big ones yet. Nobody knows where any of this goes next.
Qwen3.8-27B
Qwen3.8 Flash-Next
GLM-5.3-Flash
DeepSeek V4.1 Flash 1017
Grab the abliterated quants too while every link still works.
RIGHT NOW!
I am one of the people behind the 6,500 uncensored models the press now wants off Hugging Face.
Transformer News counted over 6,500 abliterated, uncensored and heretic models on Hugging Face in July. Their story is that these models could help cyber attacks. Hugging Face removed 1 model marketed for offensive cyber, and the reporter asks why the rest are still up.
A few of those are mine. I upload abliterated Qwen and GLM quants. My Qwen3.8 Flash-Next one got about 358,000 downloads in the last 30 days.
Abliteration removes the refusal direction. That is all. The model does not learn hacking from it. It stops lecturing you when you ask about a chemistry exam, a pentest you are paid for, or a dark scene in a novel.
The base models are already on millions of drives. Removing my repo removes nothing.
The real problem is somewhere else. Hugging Face is where almost everyone gets weights. Nvidia agreed to buy it, Nvidia just lost the CEO who promised to keep it open, and now the press asks why uncensored models are still there.
Download what you need. TODAY!!!
Dimitris Kalatzis retweeted
🚨DeepSeek居然把他们整个App也开源了。。。。
刚刚 @deepseek_ai 正式发布全家桶应用API,直接把DeepSeek App里几乎所有的功能都开放给开发者了。
包括联网搜索、深度思考搜索、文件解析、阅读链接、语音交互,以及OCR、ASR、TTS等底层能力。
甚至官方还直接开源了一套完整的聊天前端Demo,支持导入DeepSeek历史对话。
也就是说,现在开发者拿着DeepSeek的API,基本就能1:1复刻一个DeepSeek App。
DeepSeek在公告最后还专门表示:「DeepSeek现有的对话产品只是较为基础的一种形态。现在,官方产品背后的各项能力均已支持通过 DeepSeek API 随心调用、自由组合,我们期待开发者能够借此省去重复的基础工作,打造出更多样、更极致、更具想象力的产品。欢迎使用 DeepSeek API,尽情创造。」
现在各家AI公司都在拼命做自己的超级入口,想方设法把用户留在自家App里。
DeepSeek倒好,直接把自己App背后的能力全部开放出来,连复刻方案都给你准备好了。
你想做一个自己的DeepSeek?可以。
你想把DeepSeek的搜索、语音、文件处理能力拆出来,重新组合成一个新产品?也可以。
甚至可以在DeepSeek已有的能力之上,做出比官方App更复杂的产品。
这才是这次发布最值得关注的地方,DeepSeek对自己在AI生态里要扮演什么角色,似乎想得很清楚。
就是下次能不能别在北京时间的周五晚十点发布了。。。。
Dimitris Kalatzis retweeted
British police arrested a man after Facebook reported a picture he sent privately to his wife through Messenger. Police seized both their devices.
The man had downloaded an image of a Mexican actress from Google or Bing Images (he says he doesn’t remember which engine), unaware that it was AI generated. He sent it to his wife to ask if she’d like him to buy her similar clothes.
According to his account, police said the image matched a hash associated with a real Mexican actress.
If the image was illegal and could be identified through its hash, it shouldn’t have been available on Google or Bing Images. They should have remove it and report it to the police.
Anyone in the UK can face arrest, police investigation and seizure of their devices over something they download from the internet or say in a private conversation, even when they have no idea the content could be considered illegal.
Facebook Messenger is supposed to protect private conversations through end-to-end encryption. But if Meta can scan messages on people’s devices or inside the app before encryption or after decryption, that privacy protection becomes worthless against Meta’s monitoring.
The same privacy and personal safety problem applies to the ChatGPT integration for Apple Messages and Meta’s Muse spyware agent. Giving these companies access to private conversations allows them to monitor and analyse sensitive information about everyone involved, including people who never consented.
End-to-end encryption doesn’t protect people when they add third party services.
Shockingly dystopian.
Saves a pic from Google Images that was AI generated (not of a child), sends it over Facebook messenger to his wife while they're discussing clothes to buy.
Facebook reports it to police and his devices are seized. The UK is unwell.
reddit.com/r/LegalAdviceUK/c…
Dimitris Kalatzis retweeted
Europe built Wero so we wouldn't need Visa, Mastercard and PayPal… and now the Dutch bank ASN only lets you activate it if Google approves your Android phone.
The app runs Google's Play Integrity check, so anyone on GrapheneOS or /e/OS is locked out. The bank's answer: Wero relies on the "standard security controls for regular operating systems".
So the European payment system meant to get us out of US dependence gets Google as the gatekeeper… sovereignty with an asterisk
As a biologist working in drug development, I’m sick and tired of “curing diseases” being paraded around as the one thing that makes AI worthwhile. AI has not developed a single drug. Nor has it made any unique discoveries that could directly lead to the development of one.
Dimitris Kalatzis retweeted
AI bros are like:
“AI is finding a new cancer cure weekly”
But when you press them on it’s you find out:
- it’s not a cure
- it’s not for cancer
- it’s not weekly
- it’s not actually frontier generative AI
- “finding” is an incorrect term here
Literally every word is a lie.
The differing perceptions of AI are wild. The country just voted against data centers on 40+ provisions nationwide and hates AI. (Plenty of legitimate gripes.) But most of those people are just unaware that AI is finding a new cancer cure or treatment weekly, going through decades of peer reviewed data and uncovering mountains of errors, that its correcting, throughout every academic field. Replacing billions of man hours of drudgery. Closest thing to a “miracle” I’ve ever witnessed. When I talk to people outside of tech/VC, they’re just unaware of all of it.
Dimitris Kalatzis retweeted
Oh, you've "come across" another PhD student's data? How curious. It seems to be the very same student whose data you "came across" last time.
2/2 it's too much of a coincidence, just saying.
Keep your data away from AI companies, last warning
teapawt.net/AnthropicAI/status/211…
We’re sharing how our biology research team worked with Claude to map the first 3D structure of a vault particle from bacterial life.
Vault particles are a longstanding mystery. They are barrel-shaped objects, among the largest protein structures found in our cells. They were first discovered by Leonard Rome and Nancy Kedersha in 1986, but their function remains unknown.
With Claude, we’ve found results that might bring us closer to solving that mystery.
Until now, vaults were only known in the cells of complex things like animals and plants. We have shown that similar proteins from bacteria (and the viruses that infect them) can also assemble into vaults. Finding the structures in these simpler forms of life may make it easier to investigate their function in our own cells.
For our life sciences team, the project was also a demonstration of the potential (and the limitations) of using AI to accelerate science.
In our full piece, we discuss how Claude was able to dramatically speed up some stages—manually determining the vault particle’s structure would have taken weeks to months of painstaking work, but here it took days. We also discuss how Claude made some simple mistakes along the way, and how we collaborated with our machine learning researchers to address them.
Read more: anthropic.com/research/inves…
Dimitris Kalatzis retweeted
The final goal of AI companies is to end personal computing which in simple terms means to remove all access to digital devices from genral public. You can already see how prices of laptops, phones and PCs are being artificially inflated so no one can afford them anymore.
This is the largest endeavour to move digital assets from people to the technocrats. They recognise the monetary value which internet affords to common people and want to take it away. After they've taken away all your phones & PCs they will sell you a rectangle box which would run on a cloud and would require a constant subscription, this is their long game.
Dimitris Kalatzis retweeted
🦔Sam Altman says it’s “quite terrible” for regular people to keep so much memory in their own computers. He told the Cheeky Pint podcast that home PCs only use their RAM about 5% of the time. In his view, the world’s limited supply of memory does more good in the cloud.
My Take
He said this the same day Intel blamed expensive memory for a drop in PC sales. My car stays parked for most of the day too, and I’m not selling it to rent one by the hour.
OpenAI earns its money when you rent its computers instead of using your own, so of course Altman wants the memory in his data centers. I think he worries about local AI more than he lets on, because people who run good models at home have less reason to pay for ChatGPT.
Hedgie🤗
Dimitris Kalatzis retweeted
🚨🚨Apple basically told him that Israel has either just hacked his devices or is trying to.
He made the brilliant documentary on the Pager attack in Lebanon.
They literally destroyed his home last week.
Now this.
Yesterday, I received a security notification from Apple warning me that my iPhone had been targeted by a sophisticated mercenary spyware attack, involving technology of the kind developed by companies such as NSO Group, the Israeli firm behind the notorious Pegasus spyware.
Perhaps the most alarming aspect of Apple’s warning was its confirmation that such attacks target specific individuals “because of who you are or what you do.”
This is not the first time I have been targeted by spyware. I was previously targeted with Pegasus in both 2017 and 2021.
I do not yet know who is behind this latest attack, or whether it is connected to my investigative documentary, The Trap, which I directed and co-produced. The film, available on Al Jazeera 360, exposed the details of the September 2024 pager explosions in Lebanon.
What is certain, however, is that the targeting of journalists with sophisticated spyware poses a grave threat to press freedom, the confidentiality of journalistic sources, and the very foundations of independent investigative journalism.
The guy who signed for up to 900k DRAM wafers a month, about 40% of global output, is lecturing us about DRAM allocation.
The RAM in my computer is already paid for. Running a model on it takes nothing from anyone. Your data centers are a big reason RAM prices went crazy.
By that logic nobody should buy a car or a house either. We should all just rent.
A model on my machine can't be repriced, rate-limited, nerfed, or shut off when you change your mind. My data never leaves my desk. That's control you'd rather keep for yourself.
Local inference is a revenue problem for you.
The same guys who claimed a few days ago that Qwen open-weight models plant backdoors just took Canadian taxpayer money at a CAD $35.7B valuation.
Their Artificial Analysis score? 9.
Meanwhile, you have probably never heard of Motif Technologies, a Korean sovereign AI startup. They scored 47 on Artificial Analysis with Motif 3, and their valuation is CAD $136M.
Even MiniMax sits at CAD $10B.
AI is clearly an important technology, but way too much money is being burned on absolute garbage right now.
Dimitris Kalatzis retweeted
Fuck you, @ArtificialAnlys
Your new index says Claude Haiku 5.5 is smarter than DeepSeek V4.1 Flash 1017 and GLM-5.3.
Haiku is Anthropic's small cheap model. DeepSeek 1017 is the model Cursor put on its cost frontier this month, and Cursor says no model scores higher at those prices. I run it every day on my 4x 3090. I dont believe for a second that Haiku beats it.
Look at how they got there. 60% of the index now runs on private tests nobody outside can see. GDPval-AA, SciCode and the long context test are gone. In comes a private coding benchmark from Cognition. And the open weights gap goes from 3 points to 12 in 1 release, 4x. Released open models are now 26 points behind the top.
1 day before v5, on the old version, they gave Ling 3.1 Flash 1019 a 44. Now the best released open model gets 32.
The weights didnt change. The tests did, and you cant see the tests.
This is a bought benchmark. A leaderboard where you cant check a single question, that widens the gap 4x overnight and puts Haiku above DeepSeek, is marketing for closed labs. Its not measurement. I hope at least you were paid enough money to publish such garbage. Liars.
Publish the questions or stop calling it an index.
Announcing Artificial Analysis Intelligence Index v5, featuring 2 new benchmarks and 3 upgrades. For the first time, the majority of the Index is built on private datasets
The Artificial Analysis Intelligence Index is our composite measure of model capabilities across tasks reflecting real-world usage today. v5 is our biggest upgrade since v4 launched in January. With v5, 60% of the Index is now based on evaluations with private test sets or held-out answers.
Alongside Intelligence Index v5, we’re launching the Artificial Analysis Progress Index v1, which tracks model capability over time on one continuous scale.
New to the Intelligence Index:
➤ FrontierCode 2.0: Cognition's newly-launched private benchmark of real-world coding tasks, graded on whether a repository maintainer would merge the agent's change
➤ Terminal-Bench-Science: 70 agentic scientific research tasks, run in a terminal environment
Upgraded:
➤ HLE-Diamond (replacing Humanity's Last Exam): a refined 1,000-question version of Humanity's Last Exam, split evenly between reasoning and knowledge questions
➤ AA-Omniscience v1.1: An updated dataset with saturated and public set questions removed, new questions added, and ambiguous or erroneous questions revised
➤ CritPt v2: a private, 40-question hard subset of CritPt v2, revised by the CritPt authors and run with a code-execution tool
Staying in the Index: AA-Briefcase, AutomationBench-AA, GDP.pdf and Terminal-Bench 4.0
Removed from the Index: GDPval-AA v2.1, SciCode, and AA-LCR v1.1
Key results:
➤ Claude Opus 5.5 (max) leads at 58, ahead of GPT-6 Astra (max, 57) and GPT-6.1 Sol (max, 55): Anthropic and OpenAI hold the top five models. Gemini 4 Argon (high) is next at 50
➤ The open weights gap widens from 3 points to 12: Kimi K3.1 Pro (weights not yet released) is the top open weights model at 46. Among released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) lead at 32 - 26 points behind the leader and below Claude Haiku 5.5 (max, 36)
➤ Cost per task varies by more than 13x among the leading models: of the 10 highest-scoring models shown, GPT-6.1 Terra (max) is the cheapest at $0.96 per Intelligence Index task and Claude Fable 5.1 (max) the most expensive at $13.04
“We’ve built software to track Wikipedia editors,” says the head of the ADL.
We have an investigation publishing soon about pro-Israel manipulation of Wikipedia, and he just blurts it out
Dimitris Kalatzis retweeted
And always bringing back this reminder about the Minab school bombing:
teapawt.net/HeidyKhlaaf/status/209…









































