this post was submitted on 18 May 2026
16 points (94.4% liked)
technology
24373 readers
145 users here now
On the road to fully automated luxury gay space communism.
Spreading Linux propaganda since 2020
- Ways to run Microsoft/Adobe and more on Linux
- The Ultimate FOSS Guide For Android
- Great libre software on Windows
- Hey you, the lib still using Chrome. Read this post!
Rules:
- 1. Obviously abide by the sitewide code of conduct. Bigotry will be met with an immediate ban
- 2. This community is about technology. Offtopic is permitted as long as it is kept in the comment sections
- 3. Although this is not /c/libre, FOSS related posting is tolerated, and even welcome in the case of effort posts
- 4. We believe technology should be liberating. As such, avoid promoting proprietary and/or bourgeois technology
- 5. Explanatory posts to correct the potential mistakes a comrade made in a post of their own are allowed, as long as they remain respectful
- 6. No crypto (Bitcoin, NFT, etc.) speculation, unless it is purely informative and not too cringe
- 7. Absolutely no tech bro shit. If you have a good opinion of Silicon Valley billionaires please manifest yourself so we can ban you.
founded 5 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Yeah I think it was being boosted by the Tech4Palestine org, which made me lose all respect for them.
Frankly everything about this is silly. Starting from the fact that - unless this is an op - there's no way these guys had the resources to train a new model from scratch, so this is most likely a fine-tuned, reskinned open-weights model like Qwen or Deepseek.
The "using X percent less energy" claim already smells of bullshit since they don't tell you the baseline: 16X less energy than what? Also, unless these two people somehow made massive advancements in the most crowded field of the moment, the energy savings they claim just come from the underlying Chinese models which are more efficient by themselves (so these guys would be stealing credit).
The claim that "Your data stays yours, always encrypted" is straight up a lie. Encrypted LLM inference is theoretically possible (look up Fully Homomorphic Encryption), but years away from being usable or commercially viable. In other words your data has to be decrypted before being input into the LLM, so these guys can read your chats. It's the same reason Proton caught a lot of shit when they launched Lumo, their chatbot: it's their only product that isn't end-to-end encrypted, breaking with their overall product proposition. Chatbots are a privacy nightmare if they're not running locally, and anyone telling you otherwise is a liar.
The part about "Amplifying marginalized voices" is also ridiculous... They fine-tuned it to be woke? It really sounds like the woke version of Musk playing with Grok to make it behave more like a Nazi. I wouldn't be surprised if they just gave it a system prompt espousing some vague progressivism and called it a day.
As a cherry on top, the idea of funding humanitarian causes with a product that is famously a money pit is hilarious. Every AI company is setting billions in venture capital on fire trying to make an insanely expensive to run product viable, but these guys somehow figured out how to make it sustainable and even have surplus to fund ethical causes with? Come on...
I try to give every milligram of benefit of doubt to projects that claim to be pro-Palestine, but this one is just too full of bullshit.
They say it's built on GLM-4.5 Air, which is a 106B/12B active parameter Chinese open weight model from Z.ai. It definitely requires less hardware to run than an OpenAI frontier model, but the 24Wh/query number they're comparing against for chatGPT is way higher than anything I've ever heard before. The 1.5Wh/query they claim for Thaura is also weirdly high actually.
Did they do anything to it? Who can say. It sounds like they might just be running GLM-4.5 Air straight. There are lots of places you can get that. You can run it yourself if you have an Apple silicon Mac with 128gb RAM. It's even one of the free tier models on openrouter.ai.
It's not going to be useless (I actually think the most promising future of LLMs is improvement on the low end, even a 30B can be coerced into producing useful code and is legit awesome at translation and things like image recognition), but it's not a frontier model by any stretch.