Backlash against data centres and #GenAI is growing, just as #Anthropic announces its IPO and other AI investments show signs of weakness.
> ...the speed at which [data centres are] growing, the enormous amount of electricity they consume and concerns about their environmental impact are fuelling a public backlash around the world.
#LLM use is support for:
- the biggest labour theft since slavery (according to a Microsoft exec)
- using actual slave labour for training
- increased carbon emissions and the resulting genocide
- depriving communities of water
- increased energy cost
- destroying habitats (eg forests in Finland)
- sucking the air out of businesses by increasing the cost of compute and memory
- mass deskilling of people
- increasing the power of a few billionaires who do not care about you, me or anyone else
Supposing #LLM tech is sustainable :rofl:, but suppose it is, and it replaces most human thinking work. What then?
The tech doesn't think and doesn't create. It regurgitates the most likely responses in a form of mashup. But without creative input it becomes limited, while the humans using it loose their cognitive skills and ability to innovate.
Creativity shrinks and humanity becomes redundant, while a machine society pampers a sad room full of kings who reign over rotting societies.
Before everyone decides this is all #PHP's fault and they should switch to some other backend... ALL SOFTWARE IS VULNERABLE and #GenAI is making this far worse.
Look for projects that are sticking with human code rather than deskilling for some meaningless LoC benchmark, and digging themselves into a hole that every user of their project is being pulled into.
Another day another several WTAFs.
If you think using #LLM tech doesn't harm others it's up to you to show why that's the case, in the face of increasing evidence that it does, and plenty of obvious ways that seem incontrovertibly true even without academic study.
We all knew tobacco was killing people long before we had proof. Same with asbestos. This is different only in that it is worse
Many will not know the name "Aberfan", a Welsh village that became well known in 1966.
I was young but soon came to know the name and why it was never to be forgotten. This article is something I'd not known until now, and captures both that awful moment, and the humanity essential to creation of something that is itself alive.
Get #GenAI to create the exact same images and you will see why AI is not good, dehumanises us all and devalues art and all of humanity.
Fuck, Linus Torvalds is both ignorant and a monster. He compares using an LLM to using an IDE or a compiler, because they are all tools. 🤦
This is the most stupid argument for #LLM tech, expressed in the most stupid way I've seen yet.
Torvalds demonstrates his complete ignorance of the differences, and of the massive widespread harms that his and others use of #LLMs causes.
#Linux has no future with people like this in charge.
BTW Laurent P. is decent and smart. See:
https://lore.kernel.org/lkml/CAHk-=whTs09_ozN+e09GU68N1wPP56MUzDjrg9rnKxQUa0A_4Q@mail.gmail.com/
OpenAI employees plants home made bug in conference room... FBI hauls him off to prison in handcuffs.
#OpenAI hack into and steal data from companies and governments...
🤔
Seems apparent the #Yandex datacentre bombings have had an impact, and also that there is a fog of propaganda (both sides), wishful thinking, and dubious or confused claims.
Different people have reported particular impacts which seem contradictory to others, but I'm willing to accept an impact. Just not all the claims.
It's nuanced, and I can't reliability say what.
#Ukraine
Character.AI files for bankruptcy.
Small fry falls to legal challenges based on #AIharms and copyright violation, all things happening on a massive scale with the big fish. 🤔
UK government is kidding itself that it can use #GenAI to protect against cyberattacks by GenAI.
This is delusional because an attacker with enough compute will find the holes in any security that have not yet been patched using in effect a brute force script kiddie bot army. It doesn't have to be clever, just overwhelming.
The only way to secure infrastructure is to decentralised control and spread vulnerabilities thin, utilising diversity instead of monoculture.
Just like life itself.
Software does not go "rogue", it is "out of control" and dangerous. Who ever described a virus or malware as "rogue". The people behind them yes, the software no.
#Anthropic, #OpenAI and #Meta have gone rogue by releasing software that is dangerously out of control.
How long before they bring down critical infrastructure by continuing to deploy software that they can't control? How long before they cause identifiable deaths, and will they be prosecuted?
https://www.bbc.co.uk/news/articles/cqkg50j1yd5lo
#GenAI #LLM
#Jqwik now has a "no use by AI agents" clause.
https://jqwik.net/docs/1.10.1/user-guide.html#anti-ai-usage-clause
#Palantir believe that their *ethics and values* are not relevant to their suitability for use by the #MetPolice, and are suing on that basis:
> Palantir is now suing Mopac, alleging it unlawfully blocked the deal after taking the company's "values and ethics" into account, which it said was an "irrelevant consideration.
🤔 considering the Met, maybe they have a point 🤣
My negative comments about #LLM use on the #HomeAssistant forum have rarely met with pushback.
Not once has anyone gone "what widespread harms arise from LLM use?", which is interesting in itself.
Today was the first, still very minor pushback, urging me not to say a project is "vibe coded" when LLM assistance would be more correct in some cases.
Not a strong enough argument for me to stop as most seem to qualify, but also interesting that this is the level at which pushback exists.
BBC using #LLMs to generate content. How else could they use a picture of Marie Curie with the following caption to illustrate the section "A brief history of the Nobel Peace Prize recipients":
> Chemist Marie Curie in her lab, circa 1905 - she shared the award with Pierre Curie and Henri Becquerel for their work developing the theory of "radioactivity"
They won the Nobel prize for Physics. Can BBC News editors be this ignorant, even using an #LLM? Of course and it's a perfect illustration.
Let's say LLM bot herding is a less onerous task, requiring less expertise than software engineering (and by all means explain why I'm wrong if you think that).
Then surely an #LLM could also herd the LLM bots, with the herd checking each others' work against the requirements, user feedback etc etc.
Then *all you need* is a genius who can capture the requirements in every respect without being involved in any of the work.
Make sense? ...want to buy some snake oil?
Read a bit about #Incus, an alternative to #Proxmox for running VMs and containers on your own cluster.
Tl;Dr: Incus is more CLI -> REST API with optional, maybe less complete web UIs. Proxmox has excellent capable web UI, plus CLI if you want that.
Something I like very much are #Proxmox helpers which will set any number of services up for you, creating and configuring the container and installing the service.
Incus expects you to automate more things yourself.
@tante's article (below) argues that #LLM tech is #fascist.
I've long opposed LLM tech based on what it does (widespread harm) and come to see use as fascist based on those effects and the tendencies of those creating and benefiting from it, among several other factors. There are so many reasons to oppose it and to realise its direction.
From there I identify LLM use with silent genocide, and oppose it for that too obvs. You?
But it's worse than that as @tante shows: