Interesting, but disappointing tech news. EU search engines Lilo, Ecosia and Qwant are creating new EU based search index. This is something EU defined needs.

Disappointing part is that it is heavily focused on LLMs and focuses on providing for example AI summaries.

I think even more sadder part is that is might be the right choice since a lot of people are already using LLMs instead of actual search. While I hope this project succeeds, I think it is sad that LLMs are focus of everything.

@Techaltar For Nebula Q&A What do you think about this project and do you think creating AI focused search index is a right choice?

https://www.eu-searchperspective.com/

https://staan.ai/

#ai #llm #ecosia #qwant #lilo #eu

GitHub CEO: Embrace AI or get out.

businessinsider.com/github-ceo

PS. Here’s where to go: @Codeberg

codeberg.org

(As an additional bonus, you’ll have the peace of mind of knowing you’re part of an anti-fascist not-for-profit cooperative instead of a trillion-dollar US corporation that’s helping Israel commit genocide.)

@aral @Codeberg he is a bloody idiot. He doesn’t even know what #AI is! He means #LLM. LLM generates buggy code and often doesn’t even compile. Often it is not portable. It is non deterministic. That alone should say everything. The AI craze is the tech Frankenstein and it’s a giant scam. I wouldn’t worry about what someone who clearly doesn’t know what he’s talking about let alone his arse from his elbow is talking about. AI my arse.

Developing our position on #AI - by Nicholas Bergson-Shilcock (Recurse Center - July 2025):

https://www.recurse.com/blog/191-developing-our-position-on-ai

"When interacting with another person at RC it’s likely the case that both people benefit from the time spent together [...]. I think when interacting with an #LLM, depending on how they’re used, one person can scratch some social itch, and they might benefit from it, but the energy goes kind of into a technological void, instead of into the community where it benefits others."

Running Deepseek R1 671b fully locally on a $2000 EPYC server. Idle wattage is just 60w while with 260w under load.

This setup runs a 671B model in Q4 quantization at 3-4 TPS, running a Q8 would need something beefier. To run a 671B model in the original Q8 at 6-8 TPS you'd need a dual socket EPYC server motherboard with 768GB of RAM.

The idea that LLMs use an inordinate amount of power to run is very much outdated at this point.

digitalspaceport.com/how-to-ru

I just asked #CoPilot to correct spelling and grammar mistakes and to suggest improvements in an English text.

I looked at the result and could see any changes.

I asked CoPilot to highlight the changes. It did. 11 of them.

I checked again. It turned out 9 of the 11 instances were no changes or corrections (one was a spelling correction, and one was inserting the full name of a person).

This #LLM stuff seems to get worse even for my simple glorious autocomplete task 🤬

#enshittification

I don't know much about modern data centres, beyond a basic understanding of server hardware. Their penchant for water usage for because of cooling, right?

If cost were not an issue, how bad for the environment would an LLM prompt still be if fulfilled by data centre that is powered exclusively by solar/wind power, and cooled with closed-loop cooling?

#LLM#Datacenters#Environment

I still have a few Google Nest speakers around, but I hardly ever use the voice assistant. I was talking to my kid about Spider-Man and decided to ask, “How many Spider-Man movies was Andrew Garfield in?” because I forgot the titles, and if there were two or three.

It showed on the screen: “At least 24,” and then told me out loud, “55.”

It turns out he has been in two.

ETH Zurich and EPFL will release a large language model (LLM) developed on public infrastructure. Trained on the “Alps” supercomputer at the Swiss National Supercomputing Centre (CSCS), the new LLM marks a milestone in open-source AI and multilingual excellence.

"The model will be fully open: source code and weights will be publicly available, and the training data will be transparent and reproducible, supporting adoption across science, government, education, and the private sector. This approach is designed to foster both innovation and accountability.

A distinctive feature of the model is its capability in over 1000 languages. “We have emphasised making the models massively multilingual from the start,” says Antoine Bosselut.

Training of the base model was done on a large text dataset in over 1500 languages — approximately 60% English and 40% non-English languages — as well as code and mathematics data. Given the representation of content from all languages and cultures, the resulting model maintains the highest global applicability."

https://ethz.ch/en/news-and-events/eth-news/news/2025/07/a-language-model-built-for-the-public-good.html
#llm #ai #switzerland

If you have just joined mastodon, a) welcome, and b) the very first thing you need to do is add data to your profile that makes it clear you are an actual person.

I am sorry that we live in this time, none of us want this, but sometime it just be's that way.

We are all hounded by bots, scammers, spies-for-cause, and spies-for-money, all of the time. We simply cannot assume good faith from anyone who has left no trail, or who has left a trail they are not willing to share.

@GeePawHill I'm worried this will encourage oversharing increasing #surveillance for activists, and in this political climate we should all aspire to be surveillance targets, 2ndly it encourages vendor lock-in to specific #mastodon instances at least until profile migration is added and 3rdly, from what I've seen of agentic #LLM s they probably alr can create a convincing mastodon post history if you exclude personal photos and videos which again can both be bad for #privacy #deadinternet
@giacomo

No compression algorithm can decompress your file into a poem about the complicated romantic relationship between a nematode and a rosemary bush.

Beyond "lossiness", it's the user-provided context and the prediction based on that context and their own generated text that makes these models functionally useful.

Predictive text generation algorithms are not a "weapon of oppression". They are being instrumentalized to augment power, but that is not quite the same thing.

@eloquence@social.coop
No compression algorithm can decompress your file into a poem about the complicated romantic relationship between a nematode and a rosemary bush.
Why not?

If the original file contained poems about complicated romantic relationships, and texts about rosemary bushes and about nematodes, a section of the compressed high dimensional matrix of frequenced can be extracted to maximize its statistical correlation with the prompt vectors.

That's exactly what happens in any #LLM: given a lossy compression of a high dimensional frequency matrix, the software traverse and decompress unrelated fragments of the source texts according to their statistical correlation with the prompt (and the previously generated tokens).
Beyond "lossiness", it's the user-provided context and the prediction based on that context and their own generated text that makes these models functionally useful.
Plausible, not useful.

You can use them if all you need is to fool people about their interlocutor, for example to spread disinformation or win the imitation game, but that's all.

Whenever you need something correct, not just plausible to an uninformed human, they stop being useful. In fact a recent #OpenAI study recognise an error rate > 90% for its LLM on simple verifiable answer to questions.
As always, OpenAI is trying to set up a benchmark it can easily cheat, but the numbers are clear: even on basic tasks, #GenAI is totally unreliable.
Predictive text generation algorithms are not a "weapon of oppression".
The technologies in itself (the algorithms described in papers and textbooks) are not a "weapon of oppression", but all of real world models are, no matter how you use them.

So if you build your LLM from scratch properly collecting and selecting all of the source texts, you might get a model that is not harmful to people.
But if you hope to just use models from huggingface "for greated good", you are fooling yourself.
@eloquence@social.coop

While I welcome your experiments like any other attempt to challenge the power structures of #capitalism.

However this specific attempt looks pretty naive and deeply misguided.

#AI is just opaque, statistically programmed, software.

As such, it embodies and reproduce the will of those who created it, selecting the source data (instead of writing the source code). They are always special-purpose software sold as general purpose.

In particular, #LLM are lossy compressions of source data selected for the sole purpose to fool people's minds and alienate them.

In general, wielding weapon of oppression against oppressor is a wild illusion that streghten their power as even the oppressed start relying on the infrastructure of their own oppression.

And unless you compile your own source dataset from scratch, without using pretrained models from third parties, you are going to serve the very same interests and power structures you are hoping to fight.
Greg Lloyd
Greg Lloyd boosted

Jessee Bundy of the Creative Counsel Law firm pointed out that lawyers like her had been warning "for over a year" that using ChatGPT for legal purposes could backfire spectacularly.

"If you’re pasting in contracts, asking legal questions, or asking [the chatbot] for strategy, you're not getting legal advice," the lawyer tweeted. "You’re generating discoverable evidence. No attorney-client privilege. No confidentiality. No ethical duty. No one to protect you."

"It might feel private, safe, and convenient," she continued. "But lawyers are bound to protect you. ChatGPT isn’t — and can be used against you."

When an AI defender came out of the woodwork to throw hot water on her PSA, Bundy clapped back.

"I think it is both, no?" needled AI CEO Malte Landwehr. "You get legal advice AND you create discoverable evidence. But one does not negate the other."

"For the love of God — no," the lawyer responded. "ChatGPT can’t give you legal advice."

"Legal advice comes from a licensed professional who understands your specific facts, goals, risks, and jurisdiction. And is accountable for it," she continued. "ChatGPT is a language model. It generates words that sound right based on patterns, but it doesn’t know your situation, and it’s not responsible if it’s wrong." #LLM#AI
https://futurism.com/chatgpt-legal-questions-court

Jessee Bundy of the Creative Counsel Law firm pointed out that lawyers like her had been warning "for over a year" that using ChatGPT for legal purposes could backfire spectacularly.

"If you’re pasting in contracts, asking legal questions, or asking [the chatbot] for strategy, you're not getting legal advice," the lawyer tweeted. "You’re generating discoverable evidence. No attorney-client privilege. No confidentiality. No ethical duty. No one to protect you."

"It might feel private, safe, and convenient," she continued. "But lawyers are bound to protect you. ChatGPT isn’t — and can be used against you."

When an AI defender came out of the woodwork to throw hot water on her PSA, Bundy clapped back.

"I think it is both, no?" needled AI CEO Malte Landwehr. "You get legal advice AND you create discoverable evidence. But one does not negate the other."

"For the love of God — no," the lawyer responded. "ChatGPT can’t give you legal advice."

"Legal advice comes from a licensed professional who understands your specific facts, goals, risks, and jurisdiction. And is accountable for it," she continued. "ChatGPT is a language model. It generates words that sound right based on patterns, but it doesn’t know your situation, and it’s not responsible if it’s wrong." #LLM#AI
https://futurism.com/chatgpt-legal-questions-court