Available now! Telegram Research 2025 — the year's key insights 
Axis of Ordinary
Open in Telegram
Memetic and cognitive hazards. Substack: https://axisofordinary.substack.com/
Show more3 787
Subscribers
No data24 hours
-107 days
+5230 days
Data loading in progress...
Similar Channels
Tags Cloud
Incoming and Outgoing Mentions
---
---
---
---
---
---
Attracting Subscribers
June '26
June '26
+38
in 10 channels
May '26
+171
in 23 channels
Get PRO
April '26
+114
in 19 channels
Get PRO
March '26
+63
in 10 channels
Get PRO
February '26
+67
in 9 channels
Get PRO
January '26
+56
in 7 channels
Get PRO
December '25
+44
in 3 channels
Get PRO
November '25
+39
in 7 channels
Get PRO
October '25
+61
in 20 channels
Get PRO
September '25
+46
in 9 channels
Get PRO
August '25
+33
in 9 channels
Get PRO
July '25
+66
in 16 channels
Get PRO
June '25
+49
in 8 channels
Get PRO
May '25
+64
in 9 channels
Get PRO
April '25
+65
in 14 channels
Get PRO
March '25
+109
in 6 channels
Get PRO
February '25
+80
in 12 channels
Get PRO
January '25
+70
in 8 channels
Get PRO
December '24
+63
in 11 channels
Get PRO
November '24
+80
in 10 channels
Get PRO
October '24
+94
in 14 channels
Get PRO
September '24
+107
in 16 channels
Get PRO
August '24
+99
in 18 channels
Get PRO
July '24
+227
in 14 channels
Get PRO
June '24
+85
in 16 channels
Get PRO
May '24
+119
in 14 channels
Get PRO
April '24
+110
in 25 channels
Get PRO
March '24
+193
in 41 channels
Get PRO
February '24
+204
in 30 channels
Get PRO
January '24
+800
in 42 channels
Get PRO
December '23
+216
in 27 channels
Get PRO
November '23
+232
in 60 channels
Get PRO
October '23
+154
in 13 channels
Get PRO
September '23
+156
in 0 channels
Get PRO
August '23
+209
in 0 channels
Get PRO
July '23
+302
in 0 channels
Get PRO
June '23
+95
in 0 channels
Get PRO
May '23
+243
in 0 channels
Get PRO
April '23
+127
in 0 channels
Get PRO
March '23
+236
in 0 channels
Get PRO
February '23
+36
in 0 channels
Get PRO
January '23
+40
in 0 channels
Get PRO
December '22
+30
in 0 channels
Get PRO
November '22
+36
in 0 channels
Get PRO
October '22
+40
in 0 channels
Get PRO
September '22
+23
in 0 channels
Get PRO
August '22
+35
in 0 channels
Get PRO
July '22
+36
in 0 channels
Get PRO
June '22
+43
in 0 channels
Get PRO
May '22
+63
in 0 channels
Get PRO
April '22
+526
in 0 channels
| Date | Subscriber Growth | Mentions | Channels | |
| 17 June | +2 | |||
| 16 June | +1 | |||
| 15 June | +2 | |||
| 14 June | 0 | |||
| 13 June | +1 | |||
| 12 June | +2 | |||
| 11 June | +10 | |||
| 10 June | +5 | |||
| 09 June | +5 | |||
| 08 June | +1 | |||
| 07 June | +1 | |||
| 06 June | +1 | |||
| 05 June | +2 | |||
| 04 June | +2 | |||
| 03 June | +2 | |||
| 02 June | +1 | |||
| 01 June | 0 |
Channel Posts
Sam Altman: By September of this year, we will use 500,000 A-100 equivalent GPUs as an AI research intern. By March of 2028, we will have a full end-to-end, very talented researcher, like figuring out complete new architectures.
Source: https://www.youtube.com/watch?v=F_7M4Hc-usM
| 2 | Claude produces more actions (left bar) and text output per prompt (right bar) for more expert users.
https://www.anthropic.com/research/claude-code-expertise | 245 |
| 3 |  Today, we enable AutoResearch in the physical world for the first time! Introducing ENPIRE: we give 8 Codex agents a fleet of robots, an allocation of GPUs, and generous token budget. We set them free with a simple goal: solve the task as quickly as possible, keep the robots busy but stay safe, don't waste precious compute. Make no mistake.
Then humans step aside and our watch begins. The robot fleet starts to come alive: they learn to look for visual clues, reset the scene, practice novel skills, tinker with control stack, read papers online, debate, reflect, get stuck, and try again directly on the hardware. All we did is to give Codex an API to the world of atoms, and the rest is emergence.
ENPIRE is able to solve high-precision tasks like tying zip-ties, organizing fine pins, and installing GPUs all by itself. We also discovered a new type of "physical scaling": 8 robots exploring in parallel improves significantly faster than fewer ones.
Project page: https://research.nvidia.com/labs/gear/enpire/ | 455 |
| 4 | In this preregistered study, previous frontier AI models like Claude Opus 4.6 out-persuaded expert humans in conversation - incl. world-champ debaters and professional canvassers. This held even when humans chose their topics, prepared in advance, and competed for £1,000 prizes.
The AI's advantage also extended to high-stakes real-world action: The authors pitted AI against a professional fundraising firm with 7 years' experience raising money for Save the Children, to raise real donations. AI's effect on giving was nearly 3x the firm's.
The paper reports four preregistered experiments with 18,978 conversations from 6,923 persuadees.
Read the paper: https://arxiv.org/abs/2606.16475 | 1 568 |
| 5 |  Source (pixl8studio): https://instagram.com/pixl8studio/ | 434 |
| 6 |  The recent Fable order appears to ban foreign nationals from working with the most advanced AI models.
Below is the author list + place of birth for perhaps the most influential AI paper of the century, "Attention is All You Need," which introduced the transformer.
via @_NathanCalvin
[This extreme form of censorship will only become sustainable when the models become good enough to automate AI research. And then the human era will be over anyway.] | 483 |
| 7 |  RL-trained drone-racers outperform expert human pilot: https://rpg.ifi.uzh.ch/marl/
They tested out their systems in a real-world test, where the system generalized well and effectively beat the human player.
Physical deployment of our multi-agent framework is validated through racing experiments spanning time trials, AI-only races, and mixed human-AI competitions against Marvin Schaepper, five-time Swiss national drone racing champion.
Our agents outperform a champion-level human pilot in multi-player races at speeds exceeding 22 m/s, while simultaneously reducing collision rates by 50 % compared to state-of-the-art single-agent baselines. Crucially, training with diverse artificial agents enables zero-shot generalization to safer human interaction.
Through competitive self-play, anticipatory behaviors emerge without explicit programming.
Trained for “5,500 iterations, totaling 200 million environment interactions, requiring approximately 27 hours of wall-clock time on a single NVIDIA RTX 4090 GPU”. | 486 |
| 8 | Anthropic's Claude Fable 5 archives an insane score on ProgramBench: https://www.vals.ai/benchmarks/programbench
ProgramBench asks models to rebuild complete command-line programs from scratch using a compiled binary plus documentation/specification, then checks whether the produced implementation compiles and passes hidden behavioral tests.
More evidence that people feeling brain damaged after Fable has been censored by the US government are not hallucinating. | 538 |
| 9 | There are many politically convenient, yet fake, explanations for why Europe doesn't have its own state-of-the-art AI models.
Consider where AI is predominantly developed in the Western world: the San Francisco Bay Area and London, the most important non-U.S. hub. What do these places have in common? They are both extremely diverse, left-wing areas with a high degree of regulation.
Now, a bunch of OpenAI's top researchers are from Poland. Why would they leave Poland for Commiefornia? Hint: it has nothing to do with Poland being left-wing, overregulated, or suffering from high crime. Because none of this is the case.
So what is the problem? Maybe it's because German electricity prices are crazy high, so data centers cannot be built here? Well, the big US labs don't build their data centers in San Francisco but in places like Texas. The same would be possible in Europe. You could have your researchers sitting in Berlin and have your data centers in places like Finland, which are pro-nuclear and have cheap energy. Indeed, Finland explicitly markets the country for data centers because of its cool climate, affordable clean energy, abundant water, and strong digital infrastructure.
The real problem is the structure of European capital and risk-taking. Specifically, there is a lack of venture capital and a lack of willingness to pay high salaries and take on debt. Europe is much weaker at turning the resources it has into large, risk-taking, well-capitalized companies that can pay world-class salaries. The problem is that much less of Europe's wealth flows into high-risk, high-upside technology bets. Reuters recently reported estimates of Europe’s annual investment gap as high as €1.4 trillion.
If we in Europe don't quickly become comfortable with spending hundreds of billions of Euros per year on risky endeavors, we'll lose everything. | 603 |
| 10 |  Cool. But let's not deceive ourselves. China isn't going to grant access to its future state-of-the-art models that can accelerate AI research, scientific progress, and economic growth, not to mention impact their national security. Not even China's allies will have access to such models. One must either achieve AI independence, become a vassal, or be rendered utterly irrelevant.
You don't need the government for this. Even in a completely free market, you wouldn't get access to such models. We have strong evidence of this:
1. Google used AlphaChip to design the next generation of its TPUs. However, they only released it two years later.
2. AlphaEvolve, which helped Google save millions of dollars by recovering worldwide computing resources and speeding up the Gemini kernel, is still being withheld two years later.
3. The OpenAI model that solved a prominent 80-year-old math conjecture has also not been released to the public. | 603 |
| 11 |  French National Rally (right-wing) President Jordan Bardella.
The problem is that it will be impossible to tell the French people that they need to spend hundreds of billions on AI and still get elected. The average (French) voter doesn't get it. And even if they were to cooperate with Macron to agree on this issue, the far-left would ruthlessly use it against them, win, and destroy France.
It's a really sad and almost hopeless state of affairs. | 584 |
| 12 |  +3Every week now some Russian city is suffering from oil rain because Putin seeks to capture poisoned fields and rubble that will benefit nobody in Russia.
And Ukraine is forced to sacrifice its future to stop this delusional fool from sending their children to fight in his next war.
Meanwhile, people in Silicon Valley and Shenzhen are working towards capturing whole galaxy clusters. | 578 |
| 13 | No text... | 1 |
| 14 | every country should probably try and either work towards a new ai security pact with the americans immediately or pool every ounce of national resources to try and create their own ASI labs lest you become complete intellectual, economic, and moral vassals to the united states of america and the output byproducts its ASIs (you wont even get to talk to them).
– roon: https://x.com/i/status/2065939227167392147
Yes. Absolutely. But normie voters and European politicians won't be able to grasp this before it's too late (it might already be too late). Even many people smart enough to technically grasp it won't draw these conclusions because they trigger the absurdity heuristic. And even those who are both intelligent and rational enough to overcome this will have overwhelming incentives to not bite the bullet. Because biting this bullet might destroy their whole worldview and undermine the perceived importance of their favorite pet issues. | 597 |
| 15 | Amazon CEO’s Talks With U.S. Officials Triggered Crackdown on Anthropic Models
https://www.wsj.com/tech/ai/amazon-ceos-talks-with-u-s-officials-triggered-crackdown-on-anthropic-models-dcc90578 | 654 |
| 16 |  They are located in Budapest, Hungary. So this will be the end of their Fable-powered research.
Fun-fact: Even people like Andrej Karpathy, who work for Anthropic, will not be able to access Fable anymore because they are not US citizens.
Another complication with America cutting off the rest of the world from accessing SoTA models is that 60% of Google’s AI talent is located in the London.
Google hasn't caught up with Anthropic and OpenAI yet. However, if they do, issuing an executive order to limit access to U.S. nationals within the U.S. would essentially nuke Google. They are not going to move to the US either because staying in London was the deal when Google bought DeepMind.
This could significantly slow down AI research in general because it undermines IPOs and affect other AI stocks, as it makes nationalization more likely.
Even if the current orders are taken back, they're just postponed. Nobody is going to grant their competitors access to technology more powerful than thermonuclear bombs. | 707 |
| 17 | Few people realize the implications of American companies having access to a “country of geniuses in a datacenter” while we do not. That would be the end for us. None of our companies would remain competitive. It would lead to a total economic collapse.
And just to be absolutely clear about this. I'm extremely fed up with people who laugh this off while Field Medalists are already using AI in their daily research. You retards can fuck off and block me. | 700 |
| 18 |  Predictable. I didn't expect it to happen so soon, though.
Don't forget that in a few years, the United States will have models that are much smarter than Fable.
This is a wake-up call for Europe. We need to immediately invest at least two trillion dollars in building data centers, developing frontier models, and constructing power plants. We need to use our leverage with ASML and Zeiss to buy chips for as long as possible. Otherwise, we're truly fucked.
Of course, this won't happen. It's over.
https://x.com/AnthropicAI/status/2065597531644743999 | 883 |
| 19 | Russian asset Tulsi Gabbard released a map of bioweapons labs in Ukraine that has Kyiv right next to Moldova.
https://www.dni.gov/files/BIOLAB_Slides.pdf
It would be funny if it wasn't so sad. | 709 |
| 20 | Claude Fable 5 scores 88% on Tier 4 of the FrontierMath benchmark. It consists of research-level mathematics problems intended to be hard even for professional mathematicians in the relevant field.
More: https://epoch.ai/frontiermath/tiers-1-4 | 717 |
