Placing Ghosts into Shells.
HBM
Joined February 2022
- Tweets8.8K
- Following668
- Followers382
- Likes24.2K
I have been thinking about this a lot since the days we got reasoners on my blog, back when training was more open, and have yet to find credible evidence of emergence of said general intelligence that cannot be explained by combinatorial induction from training data.
Arguably the neuralese we get from capable large models can be considered a novel encoding (non-combinatorial innovation) and we are still ~1 OOM away from human synapse ranges, even in 10T models, but each time i look inside generality claims, a human demonstration, preference or prior is looking back at me in the training data.
While most of us lament the downfall of european industry and pretened aliens going to pay for our social economies, mistral here uses their investor money to deploy GPU clusters in nuclear powered France.
Prices on chips are up by over 170% just this year, wait times are 36+ weeks and neoclouds love to squeeze any model lab on a whim.
Mistrals models are still catching up for now, but this is the foresight that wins the long term.
ML4 was trained on 3,800 NVIDIA Grace Blackwell GPUs in our European datacenters -- our cluster in Bruyères-le-Châtel built using our Series B fundraise.
We are investing heavily in infrastructure, with our Series C and D clusters coming online soon. This additional compute will let us train longer, pursue more ambitious post-training, and iterate faster.
We expect large and rapid improvements in the weeks and months ahead.
2/n
It is clear to me none of the EU lawmakers use any AI if they think chatgpt needs watermarking
‼️ BREAKING: OpenAI will start invisibly watermarking ChatGPT and Codex text in the EU over the coming weeks to comply with the EU AI Act.
The company admits the watermarks are "often undetectable, especially in short passages," and that rewriting or translating can remove them completely.
Only approved researchers get the detector.
Tensor Templar retweeted
i'm realizing now that this has been simmering in me for a while.
i'm fed up of hearing people react to engineering failures with eschatological hysteria. i'm tired of labs emphasizing the fear.
look, i get it, things can go wrong. but you've demonstrated nothing but engineering problems. now either put on your big boy pants and find a way to fucking fix them, or get a different job, because the world needs this technology.
I am a huge fan of simple e2e training recipes and think this work belongs on the must read list for pretraining. It is the SSTO of LLMs
Tensor Templar retweeted
We did it: The EU Parliament continues to OPPOSE #ChatControl 2.0! 🥳
Thanks everyone for writing to representatives. 🫶
The next trilogue is scheduled for November, we must make sure that end-to-end encryption and private communication remains to be protected!
#privacy
🚨🇪🇺 The #ChatControl trilogue starts on Tuesday & the EU Council is trying to pressure Parliament into giving up its opposition. We must stop them once again! 💪🏼
Ahead of tomorrow’s final CSAR trilogue, leaked Council documents suggest a proposed compromise around so-called "search plans". These would allow authorities ask providers to scan broad groups of users or entire services for extended periods.
This isn't targeted policing, it is mass surveillance!
It's crucial that the EU Parliament remains strong and keeps up their opposition.
Because no matter the framing, the truth remains the same: There's no backdoor for the good guys only!
We at Tuta will continue to fight for your right to #privacy. Make your voice heard as well, today is the day!
➡️ fightchatcontrol.eu/
Ambitious, i like it. Launch costs are why this is not going to be a competitor unless they have a reusable rocket factory to unveil next.
A Starship test flight added 20tbps in a single launch.
We’re excited to welcome @animesh_garg as our new Chief Research Officer at @microagi, with the team from @usablerobots joining him.
He has spent most of his career working on one of the hardest problems in robotics: getting robots out of the lab and into the real world.
Frontier AI now works. Hardware is becoming cheap. And the industry is finally learning how to collect robot data at scale.
What’s still missing is post-training.
Taking a general-purpose model and teaching it one specific job, in one specific environment, until a company can actually depend on it.
In LLMs, post-training is what turned impressive demos into products. In robotics, that layer barely exists yet.
A strong model is a capability. A robot that can perform the same job, on the same line, every shift, without an engineer standing next to it, is a product.
And that process has to happen again for every task, every site, and every customer.
We believe the company that can do this well, fast, and at scale will be the company that puts robots into production at massive scale.
MicroAGI is building both sides of that system.
With @joinshift, we’re building one of the most diverse and largest egocentric datasets in the world.
With Atlas, we post-train and harness models directly on enterprise deployments.
Our goal is simple:
Deploy 1 million robots.
Tensor Templar retweeted
STOP CHAT CONTROL 2.0. - What you should know:
Tomorrow, 29 September, behind closed doors at the Council, the most draconian version of Chat Control 2.0 goes on the table. It is no longer only an intermediary reading all your private messages and cloud storage. It is a licence to search them for months, without suspicion against any individual.
The leaked Presidency note to delegations (11956/26, 10 September) proposes "authorisation of search plans for a period of time rather than on a case-by-case basis" for private chats and emails.
In plain terms: your messages, contacts, photos and cloud will be scanned under a state-issued licence for months, with no suspicion against you.
The same note admits that under this regime "a high volume of reports generated do not lead to any operational outcome." The Council's Legal Service already warned this is general and indiscriminate surveillance, illegal under EU law. The @Europarl_EN mandate says the opposite: judicial authorisation, targeted to specific persons, no scanning licences.
Tomorrow the pressure is on Parliament to fold. You remember the tricks used to push Chat Control 1.0 through and how @EPPGroup with @RobertaMetsola helped to push for it.
So fasten your seat belt for what is coming tomorrow, and call or write to your MEP and your government now to reject Chat Control 2.0. as a whole!
Remind them what this does to the privacy of every one of us.
Under a search plan, every chat, photo or contact the scanner flags is stored by the provider and sent to a new EU agency, the EU Centre, which keeps it and passes it to police. Any state authority can then demand it, with a deadline and a fine for delay, and nobody has to check who is really behind the request, in case it is hacked or weaponised.
That is exactly what happened at Revolut, and what keeps happening through abusive requests from authoritarian states, as we have reported for 13 years.
In the Revolut case the hackers got access to a real state mailbox. Their requests came from an authentic certified email (PEC) account of an Italian Prefecture on the Interior Ministry's domain, with Postal Police protocol numbers, as European Investigation Orders in the name of the Milan prosecutor, for about five months.
About 680 customers lost passports, selfies, home addresses, IBANs and full transaction histories. The files are being sold; hundreds of victims now fear kidnapping. No remedies exist as of now to protect against such attacks.
Let's stop it before it is too late. Please repost.
We need more benchmarks like this. Both mecheng and EE. I am happy to buy a stm32 but what happened to all the chip/component 3D print startups?
AGENT WARS
We asked 5 top AI models to design, engineer and 3D print a bridge to with the goal of holding the most possible weight.
🇺🇸 Anthropic: Claude Opus 5.5 (High)
🇨🇳 Kimi: Kimi K3 (High)
🇺🇸 Meta: Muse Spark 1.3 (High)
🇺🇸 OpenAI: GPT 6 Astra (High)
🇺🇸 SpaceXAI: Grok 4.7 (High)
Each bridge had specific parameters. It had to span 2 feet, use a maximum 500 grams of filament, and it had to be under 18 hours of printing time. We specified what type of weight and where we would place it on the bridge.
The Score:
• Claude Opus 5.5 held an estimated 130 lbs! It took 9 hours and 11 minutes to print 17 parts and used 441 grams of filament.
• Meta Muse Spark 1.3 held 26.5 lbs. It took 13 hours and 12 minutes to print 49 parts and used 478 grams of filament.
• OpenAI GPT-6 Astra held an estimated 17.5 lbs. It took 15 hours and 44 minutes to print 29 parts and used 442 grams of filament.
• SpaceXAI Grok 4.7 did not finish. On assembly it could not stand up by itself. It took 12 hours and 22 minutes to print 29 parts and used 460 grams of filament.
• Kimi K3 did not finish. It wasn’t engineered correctly so the bridge could not even be assembled. It took 11 hours and 19 minutes to print 35 parts and used 446 grams of filament.
That means our champion for this episode is Claude Opus 5.5. Not only did it hold the most weight (by FAR) it also took the least time to print, used the least amount of parts and used the least amount of filament.
This was episode 02 of Agent Wars, stay tuned for episode 03.
Tensor Templar retweeted
Here's who made the cut for tonight's big state dinner at the White House. In the old days, when a noble House was suddenly missing from a royal function, it was a deliberate, public signal from the King that they had lost his protection. Usually your last chance to flee.
Tensor Templar retweeted
microagi x ElevenLabs: what happens when you work together with a robot?
Language has always been at the heart of human connection. It's how we share ideas, build communities, and understand one another. We believe this will be just as important in how we connect with robots. At microagi, we envision a world where robots work alongside us as a normal part of everyday life, where asking one for help feels as natural as asking the person beside us.
For that to happen, we need to be able to speak to them directly, without a screen or control panel, and trust that we have been understood.
With ElevenLabs, we are working to make those conversations possible. We are giving our robots a voice, and shaping how they listen and respond to the people around them.
Here’s an early look at how future robots could talk with us.
"People aren't ready, this is gonna change everything"
You know what people weren't ready for? Fucking mammoths. Tigers eating them. Monkeys ripping their arms off. The black plague. The ice age. Smallpox. Vikings pulling up on the beach. Drinking water that gives you diarrhea until you die. Getting a scratch and losing your whole leg. The Mongols. The Spanish Inquisition.
One volcano can send us back to the dark ages and these safety guys are gonna be shocked. Oh the robots didn't kill us? It was a volcano? Imagine their disappointment as the hot ash rains down on them.
A solar flare is gonna explode your phone in your pocket your legs are gonna be covered in battery acid and your bank account will stop working. You're afraid of a computer program doing math?
"Things are moving too fast" homie you live on a rock that periodically deletes everything. You are one the most luckiest entities in a 100 light year radius living during a golden era. Cheer up.
My ancestors crossed an ocean in a wooden box because they ran out of fucking potatoes.
"The social disruption will be enormous!"
Nigga World War 2? World War 2? Hop in the people carrier we're landing on the beach.
Are these people out of their fucking minds? Do they not know what we've survived already?
A couple decades of air conditioning and people suddenly think oh this is the natural condition of mankind. No. We used to have eight kids because seven didn't make it. You're gonna be fine. Man the fuck up.
Accelerate.
