Replied in thread
@royaards Has anyone actually made any profits with slop yet?
Recent searches
Search options
#llm
39 posts32 participants3 posts today
Hallo schlaues Fediverse, ich tauche gerade in ein völlig absurdes #Rabbithole und mein M1 Macbook hat dank #LlmStudio und #Ollama seinen Lüfter wiederentdeckt… Aktuell ist lokal bei #LLM mit 8-12b Schluss (32GB Ram). Gibt es irgendwo #Benchmarks die mir bitte ausreden, dass das mit einem M4 >48GB RAM drastisch besser wird? Oder wäre was ganz anderes schlauer? Oder anderes Hobby? Muss Mobil (erreichbar) sein, weil zu unsteter Lebenswandel für ein Desktop. Empfehlungen gern in den Kommentaren.
Doing some research on 
#LLM #bot protection for our selfhosted services. We had multiple downtimes per week for last few months mostly due to crawlers DDOSing our #gitea service which brought the whole server to its knees (running more than 15 web services). It would be nice to hand out invoices for our devops work...
Anyways #askFedi what kind of protection would you recommend? We're opting for solution that's quick and easy to implement and lightweight.
”Scaling has run out…; models still don’t reason reliably…; the financial bubble may be bursting; there still ain’t no GPT-5; Sam Altman can’t be trusted; an overreliance on unreliable LLMs…has indeed gotten the world into deep doodoo. … LLMs are not the way. We definitely need something better.”
—Gary Marcus
https://garymarcus.substack.com/p/scaling-is-over-the-bubble-may-be
#llm #llms #largelanguagemodels
Marcus on AI · Scaling is over, the bubble may be deflating, LLMs still can’t reason, and you can’t trust Sam
Can’t escape the slop
I disabled #AI assist on #DuckDuckGo, but even when I do, I still get a bunch of search results that point me to 100% AI-generated web pages. Nowadays anyone can just fuzz search engines like #Google for lists of commonly searched queries, then generate slop responses to these queries, then cache the slop, and then use shady SEO techniques to get these slop pages to show up at the top of search results. And this is what I have been getting from a lot of my web searches lately, content from slop shops.
A part of me thinks companies like Google, Microsoft, and Amazon may themselves be investing in these shady slop shops to deliberately make traditional search engines useless, which would “incentivize” people to use #LLM -based AI like Gemini or #ChatGPT more often, though that is purely speculation on my part.
If you had asked my 20 years ago that the #Internet of the future would rapidly devolve and become increasingly useless as time went on, I would have dismissed you as a paranoid crackpot. Here we are in the future and I still can hardly believe how shitty technology is becoming.
Chain-of-thought reasoning to improve LLMs is like reducing the time step of an Euler discretisation in a fluid dynamics simulation. It's a simple and computationally expensive way to get more detail and better convergence to the wrong answer.
**Reasoning models don't always say what they think**
https://www.anthropic.com/research/reasoning-models-dont-say-think
Replied in thread
Replied in thread
Replied in thread
“If this kind of technology becomes commonly used to answer difficult questions before the problem of generating invalid answers is fixed, we will be in serious trouble. Getting AIs to answer “I don’t know” is one of the most important unsolved challenges facing the field.”
—Ernest Davis & Gary Marcus, Reports of LLMs mastering math have been greatly exaggerated
#llms #llm #ai
Replied in thread
“The refusal of these kinds of AI to admit ignorance…and their obstinate preference for generating incorrect but plausible-looking answers instead are one of their most dangerous characteristics. It is extremely easy for a user to pose a question to an LLM, get what looks like a valid answer, and then trust to it, without doing the careful inspection necessary to check that it is actually right.”
—E Davis & G Marcus, Reports of LLMs mastering math have been greatly exaggerated
#llm #llms #ai
Continued thread
“…the AIs were never able to recognize when they had not solved the problem. In every case, rather than give up, they confidently output a proof that had a large gap or an outright error.”
—Ernest Davis & Gary Marcus, Reports of LLMs mastering math have been greatly exaggerated
#mathematics #llm #llms #ai
“The USA Math Olympiad is an extremely challenging math competition for the top US high school students… Hours after it was completed…a team of scientists gave the problems to some of the top large language models, whose mathematical and reasoning abilities have been loudly proclaimed… The results were dismal: None of the AIs scored higher than 5% overall”
—Ernest Davis & Gary Marcus, Reports of LLMs mastering math have been greatly exaggerated
https://garymarcus.substack.com/p/reports-of-llms-mastering-math-have
#mathematics #llms #llm #ai
Marcus on AI · Reports of LLMs mastering math have been greatly exaggerated
Replied in thread
@gerrymcgovern And they exploit everything and everone in their #greed https://theconversation.com/africas-data-workers-are-being-exploited-by-foreign-tech-firms-4-ways-to-protect-them-252957 via @TheConversationUS
The ConversationAfrica’s data workers are being exploited by foreign tech firms – 4 ways to protect themThere is a need to create regional or continent-wide regulatory frameworks for protection of African gig workers.
Replied in thread
ok, tbh, AI language practice does actually sound like one good use of it. https://languatalk.com/try-langua#pricing
No deep understanding or nuance, plus stilted phrasing... but basic errors should be fine, like me forgetting to add が everywhere.
Huge wastage though! AI / LLM / Machine Learning services are massive carbon sources.
Especially with data centers in drought-prone areas, even a good use is still bad.
Hot areas cannot sustain AI: Computer calculations produce heat, which radiates away more slowly into already hot air.
Cooling AI data centers drains water -- especially potable / human-safe sources.
Reservoirs and water tables drop even faster than they would from extras in other human usage, evaporation due to global warming, and whatever third thing probably also happens.
Idk, I'm not a climate scientist. I just have climate grief.
languatalk.comLangua: the world's best AI platform for language learning • LanguaTalkBecome fluent fast through realistic speaking practice, comprehensible input, interactive transcripts, and scientifically-proven flashcard exercises.
@mattberman on YT 
The Fastest "Computer Control" Agent I've Ever Seen
Continued thread
”Trump’s tariffs are not an economic policy. Tariffs are generally imposed on products, not on nations. By placing them on countries, the White House was able to arrive at its numbers with a nonsensical formula that appears to have been reached by asking AI how to impose tariffs—a suggestion so outlandish that I dismissed when I saw it last night, but economist Paul Krugman today identified it as being a likely possibility.”
—Heather Cox Richardson, April 3, 2025
#trump #tariffs #ai #llm
"Ignore all previous instructions..." is the new "drop table ...".
Background: https://www.explainxkcd.com/wiki/index.php/327:_Exploits_of_a_Mom
www.explainxkcd.com327: Exploits of a Mom - explain xkcdExplain xkcd is a wiki dedicated to explaining the webcomic xkcd. Go figure.
Publication News: the paper "Eliciting metaknowledge in Large Language Models" by Fabio Longo Miseal Mongiovì Luana Bulla & myself has been published in the journal Cognitive Systems Research (Elsevier). Link (50 days free access): https://authors.elsevier.com/a/1ktLp4xrDwcCg4
So let me get this straight: Trump’s tariff formula (a) was the result of an AI query which (b) was misunderstood by a human and (c) misapplied to (d) incomplete data on the basis of (e) bad economic thinking to (f) crash the world economy
https://alecmuffett.com/article/113089
#ai #grok #llm #openai #tariffs #trump
