Is this just sloppy speaking or do people really believe that „AI understands sth and then helps you do sth“, while it rather is „the #GenAI-based system extracts features from existing data and applies them to new data“?
As educators and scientists, we can and should communicate clearly that generative AI tools are not sentient, have no capacity for truth, and are merely complex statistical algorithms dressed up in a plain language outfit.
Oh, joy, and now something inside Meta is crawling me. This makes me think that tthe #GenAI overlords are watching YCombinator and crawling everything that gets on the front page. Meta User-Agent: "facebookexternalhit/1.1 (+http://www.facebook.com/externalhit_uatext.php)"
Hmm, in the dim dark past there was a "Facebook API" and people would write "Facebook Apps", is that still a thing? Because this could be that… it really looks like a systematic (but very slow) scrape.
@timbray@me@michaelrussell Their API would allow someone to automatically post every link on your site to FB, and that would trigger the link previewer. Why someone would do that is a mystery. Maybe the bot writer is also incompetent? I get so many hits on my site that are clearly from "Baby's First Bot". Like loading URLs with their code's $variable_names non-expanded.
I see that openai.com/gptbot is crawling my blog, top to bottom, side to side. I’m sure OpenAI has consulted the “Rights” link clearly displayed on every page, invoking a Creative Commons license that freely grants rights to reuse and remix but not for commercial purposes.
“I speak to a lot of businesses around #AI, and particularly #GenAI, and I’m sensing a #hype fatigue. Part of this is due to the challenging of bridging the gap from PoC to production"
I can't imagine working without GenAI any more. I often write quick bash scripts to automate things, but for some reason, the syntax always falls out of my head and I'm constantly looking things up.
Now I just hit ChatGPT and ask it to write the script for me. With the latest version, is usually works perfectly the first time, so long as I craft a good prompt. This is a huge productivity boost.
This generative model allows you to sketch out a scene with a few words, it then leverages an LLM to flesh out the details, with the ultimate goal of feeding those details to a downstream visual image generation model.
It is almost, but not quite, entirely the inverse of image captioning models.
This offers the closest experience to an image generation tool that's usable by people with visual impairments.
"In [the counterfactual task] paradigm, models are evaluated on pairs of tasks that require the same types of abstraction and reasoning, but for each pair, the content of the first task is likely to be similar to training data, whereas the content of the second task (a “counterfactual task”) is designed to be unlikely to be similar to training data." -- #MelanieMitchell
"Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?"
To expect a priori from an LLM more than an answer equivalent to an interpolated data point as obtained from Monte Carlo data augmentation seems unwarranted.
Or likewise, to expect from a Random Forest classifier to correctly classify an input whose parameter values falls outside the learned parameter ranges.
As expected, this study shows that LLMs fail at "counterfactual" tasks – at providing answers to questions outside the training set.
"Embers of autoregression" indeed: the average person dramatically underestimates the amount of knowledge others have, collectively, and hence expresses surprise at the "right" answers an LLM spits out – failing to comprehend the sheer size of the data sets that went into training the LLM in the first place. But query outside that corpus of knowledge, even query about that corpus but in an unexpected way, and the LLM unwittingly responds with nonsense.
Question of the day. Is the whole beautiful mass of free and open Internet knowledge now to be considered as the satanic mills of AI Gen Big Tech? At their mercy, to do with as they please.
@DrPen Yes, just as with humans, why should an algorithm not be allowed to use free and open accessible data , the concentration of power in big tech is the real problem not the free accessible data they can (mis) use
@ErikJonker Yes exactly. If, for example, we had technosocial contracts between cities or nations and global tech companies to operate for the public good, then the internet of knowledge could be a vast system of interactive learning networks, delivering knowledge in different ways to different search spec's through interoperable apps. Just think what that would be like.
@denis forget who said it, but was something along lines of: “before you trust ChatGPT about something you don’t know, first ask it to answer about a topic you know a lot about”