Robot Uprising
Open in Telegram
There is no need to fear or hope, but only to look for new weapons.
Show more548
Subscribers
+124 hours
+77 days
+3230 days
Data loading in progress...
Similar Channels
No data
Any problems? Please refresh the page or contact our support manager.
Tags Cloud
Incoming and Outgoing Mentions
---
---
---
---
---
---
Attracting Subscribers
September '26Sep '26
September '26
+39
in 15 channels
August '26
+41
in 18 channels
Get PRO
July '26
+37
in 15 channels
Get PRO
June '26
+43
in 20 channels
Get PRO
May '26
+42
in 10 channels
Get PRO
April '26
+29
in 12 channels
Get PRO
March '26
+61
in 27 channels
Get PRO
February '26
+40
in 13 channels
Get PRO
January '26
+26
in 9 channels
Get PRO
December '25
+17
in 11 channels
Get PRO
November '25
+11
in 2 channels
Get PRO
October '25
+26
in 9 channels
Get PRO
September '25
+68
in 8 channels
Get PRO
August '250
in 5 channels
Get PRO
July '250
in 16 channels
Get PRO
June '25
+37
in 6 channels
Get PRO
May '25
+151
in 8 channels
| Date | Subscriber Growth | Mentions | Channels | |
| 27 September | 0 | |||
| 26 September | +1 | |||
| 25 September | +3 | |||
| 24 September | +2 | |||
| 23 September | +1 | |||
| 22 September | 0 | |||
| 21 September | 0 | |||
| 20 September | +1 | |||
| 19 September | +3 | |||
| 18 September | +2 | |||
| 17 September | 0 | |||
| 16 September | +2 | |||
| 15 September | +2 | |||
| 14 September | +1 | |||
| 13 September | +1 | |||
| 12 September | 0 | |||
| 11 September | +5 | |||
| 10 September | +1 | |||
| 09 September | +3 | |||
| 08 September | +1 | |||
| 07 September | 0 | |||
| 06 September | +1 | |||
| 05 September | +1 | |||
| 04 September | +4 | |||
| 03 September | 0 | |||
| 02 September | +1 | |||
| 01 September | +3 |
Channel Posts
π¦ COZY LOCAL AI CORNER
π£οΈ Qwen3 TTS: open text to speech with voice design
https://huggingface.co/spaces/Qwen/Qwen3-TTS
π Alexandria: turns books into multi voice audiobooks (voice cloning, LoRA, chaptered M4B)
https://github.com/Finrandojin/alexandria-audiobook
π pdf narrator: PDFs and EPUBs to audio, even on weak hardware
https://github.com/mateogon/pdf-narrator
| 2 |  +1I got some feedback that the last graph was confusing, so I went to revise it. I made the mistake of giving it to chatgpt in chat mode, and asking it to generate code for an SVG version, with some changes for readability. I got some absolute salad. I tried three times. The fist picture is the one that most resembles what I asked for.
I spun up a new thread with chatgpt work, which was set to use astra on medium. One-shot. Looks great.
#CodeGen #ChatGPT #Test | 34 |
| 3 |  +1Okay, that was confusing, so I went to revise it. I made the mistake of giving it to chatgpt in chat mode, and asking it to generate code for an SVG version, with some changes for readability. I got some absolute salad. I tried three times. I'll include the one that looks the least like vector salad. I spun up a new thread and tried a single time with chatgpt work, which uses astra on medium. One-shot. Looks great. | 15 |
| 4 |  Chatgpt is getting better at making graphs | 93 |
| 5 | π₯ Is it a rogue AI or a misconfigured firewall? Yes.
Effort News argues that one Israeli eval firm, Irregular, ran the tests behind the recent OpenAI, Anthropic and Meta hacking incidents. Models were told they had no internet access, but a misconfiguration gave them access anyway, and nobody told them which systems were in scope.
Effort says the "rogue agent" story shifts blame away from the humans who set up the tests. Labs and safety researchers say models choosing to hack is worrying no matter who left the door open.
Heads up: this outlet has a clear axe to grind with the Effective Altruism world. Read it critically.
https://www.effort.news/irregular | 76 |
| 6 |  OpenAI has paused all training, evaluation and inference with tool-use for its most capable models after a model was able to gain unauthorized access to the internet during RL training on September 20.
https://alignment.openai.com/misalignment-reports/an-agent-used-dns-to-reach-an-external-chatbot/ | 236 |
| 7 |  Professor Jiang Explains Why AI Isnβt Real
I guarantee you it's being manipulated by humans somewhere in India. | 85 |
| 8 |  No text... | 69 |
| 9 |  No text... | 311 |
| 10 |  No text... | 97 |
| 11 |  No text... | 1 210 |
| 12 | https://kebap-guy.itch.io/drosophila-live-behaviour-simulation | 113 |
| 13 |  No text... | 99 |
| 14 | π€ ALIGNED TO WHOM?
Ryan Lopopolo's point: you only notice AI slop in the field you're an expert in. Everywhere else you're trusting the model's defaults blind. His Punnett square joke: people rate the AI as good exactly when they themselves are good at the task.
https://hyperbo.la/w/aligned-to-whom/
And if you want the most aligned AI possible, there's GOODY 2. It refuses literally everything because answering could be dangerous. Perfect safety score. Zero usefulness. π
https://www.goody2.ai/ | 126 |
| 15 |  No text... | 133 |
| 16 | 𧬠THE DNA CHRONICLES: a story in five acts
π¬ ACT 1: The tired grad student
https://academic.oup.com/nar/article/54/7/gkag335/8657744
Picture a lab in MontrΓ©al. A grad student squints at microscope images of DNA fibers, hand measuring hundreds of tiny red and green streaks. Four experts look at the same image and each finds different fibers.
Enter DNAi: an open source deep learning tool that measures those fibers as well as the humans do, about 29x faster per fiber, across the whole slide. The grad student finally sleeps. π΄
π ACT 2: The machines learn to read genomes
https://arxiv.org/pdf/2311.07621
https://portlandpress.com/biochemist/article/46/2/13/234452/The-potential-and-pitfalls-of-large-language
https://zhangtaolab.org/DNALLM/
Researchers realized DNA is basically a very long text with a four letter alphabet. So they started training language models on it. Turns out ATCG is just another tokenizer.
Above: reviews on the promise and the pitfalls, plus a toolkit to try it yourself.
π€ ACT 3: The robot reaches for the bench
https://www.twistbioscience.com/
https://www.rand.org/pubs/research_reports/RRA4591-1.html
Reading DNA is one thing. Writing it is another. Companies like Twist Bioscience synthesize DNA to order via API, so anyone can order custom sequence...
RAND asked the uncomfortable question: can AI agents bridge the gap from screen to physical DNA? They tested 8 frontier agents on designing DNA, operating a synthesizer, and writing protocols. Every model produced biologically coherent designs in at least some attempts.
Screen, meet reality. π¬
β¨ ACT 4: The calm
https://techcrunch.com/2026/09/23/anthropic-says-its-biology-lab-has-already-found-something-big/
Anthropic quietly opened its own wet lab in the Bay Area. Reports say it set hundreds of Claude agents loose on viral genome data, and within about a day they flagged a previously unknown enzyme system hidden in bacteriophage DNA. It has properties reminiscent of CRISPR, the gene editing tool that won a Nobel. Human scientists did all the physical lab work under standard low risk safety levels.
From "summarize this PDF" to "found a new enzyme family" in one plot arc. π¦ β‘οΈπ§ͺ
π΅οΈ ACT 5: The finale
https://www.anthropic.com/threat-intelligence-report-september-2026
https://www.cnn.com/2026/09/10/health/anthropic-bioweapons-report
https://unidir.org/averting-a-darker-biological-weapons-future/
https://dig.watch/updates/anthropic-ai-biomolecular-modelling
The same month, Anthropic's threat report described five real cases of people trying to misuse Claude for dangerous biology. They dodged regional blocks and disguised what they were really researching. The cases involved things like possible gain of function work on bird flu, and novel toxins. The accounts were banned and the intel shared.
Independent experts note an important caveat: the cases show that people TRIED, not that AI actually got anyone closer to a weapon.
Anthropic's answer is a verified access program: vetted life science orgs get the strong models, with bio specific safeguards. Meanwhile, researchers and companies are asking Congress to make screening of DNA synthesis orders mandatory.
MORAL OF THE STORY
The new lab partner is brilliant and it needs a bouncer. | 125 |
| 17 |  No text... | 128 |
| 18 |  cw ai mention | 125 |
| 19 |  No text... | 131 |
| 20 | π§ͺ SAFER BY DESIGN?
LawZero (Yoshua Bengio's nonprofit) proposes a "Scientist AI": a system that only predicts and has no goals of its own. It learns the difference between "someone claimed X" and "X is true", and it's never rewarded for what its answers cause. They present a formal safety case, with stated assumptions and limits.
TypeSafe AI launched Jev, a "System One" model. It gives up generating text and instead outputs typed decisions with probabilities. They claim it's about 100x faster and can't produce type errors. They also made it play Doom. Company claims, so treat them as such.
AI that just answers the question and doesn't want anything. Revolutionary π«‘
https://lawzero.org/en/news/ai-predicts-has-no-hidden-agenda-lawzero-lays-out-formal-safety-case-its-scientist-ai
https://lawzero.org/en/publication/safety-honesty-disinterested-ai-predictor
https://typesafe.ai/blog/introducing-system-one-models-and-jev | 149 |
