"The central claim of the tech companies selling LLMs is that any work that people do that results in text artifacts is just “text in-text out” and can therefore be replaced by their synthetic text-extruding machines."
@baldur If #GenAI has a place in education at all, it’s to teach us some things we already should have known:
That text has no inherent value or meaning
That making a text longer (or shorter) has no inherent impact on its value
That vocabulary and grammatical correctness are tangential to value
Which means that in approximately all situations where an #LLM is used, the prompt is more valuable than the output. Once we recalibrate culturally, LLMs will be seen as worse than useless.
Trying to get my head around the GenAI APIs popping up all over lately. What's a good use-case for wrapping calls to ChatGPT in your own application?
For instance, the Spring AI's for calling various LLMs. Seems like you could make cool party tricks of creating poems, maybe creating content for your marketing, maybe images based off your data... But is any of it useful?
Feels like these marketecture articles about "10 ways to use GenAI in Banking" (or insert your domain here) can all be boiled down to "we can help you create marketing content".
In all other areas -- the downside of generating inaccurate info vastly outweighs the benefits, right? #genai
My overall impression of the utility of #genAI so far hinges on two things: 1. hallucinating is the point, and 2. how copyright claims shake out (only human-created works can claim copyright, and no, prompt engineering doesn't count)
The intersection where you want (or can tolerate some) hallucination but don't want copyright is probably pretty small.
A cybersecurity researcher finds that 20% of software packages recommended by GPT-4 are fake, so he builds one that 15,000 code bases already depend on, to prevent some hacker from writing a malware version.
Disaster averted in this case, but there aren't enough fingers to plug all the AI-generated holes 😬
This one is sooo good. I recommend this to anyone playing with #AI to understand the biases and the complexities. Oh and the discussion of alt text is amazing.
Inside LAION-5B, an AI training dataset of 5B+ images that has been unavailable for download after researchers found 3,000+ instances of #CSAM in December 2023.
I love how OpenAI's image description feature pretends to pat itself on the back when describing an image after telling me that part of an image is not clear so no text is visible. My system prompt specifically instructs it to not make things up when something isn't clear.
I haven't yet come to terms with all the crazy things that have happened in the last 4 years, but I'm now getting the feeling that I can pretty much say goodbye to over half of my friends due to differences in opinion about Covid, AI, and NFTs. On the plus side, none of it is Trump related, but I'm not yet sure about Israel/Palestine and it kind of scares me to think that even more things can come up that will further isolate me. It's crazy.
One nightmare scenario is that OpenAI and Midjourney effectively win the plagiarism battle and they’re embraced as industry standards, not because they’re right, but just because enough people were stupid enough to say it’s ok to use them.
In some ways it’s like car culture and the fossil fuel industry where despite all the harm they do to humans and the environment, acceptance by the majority basically seals our fate.
A look at Databricks' new open source #AI model DBRX, which cost ~$10M to develop over several months and, Databricks says, outshines Llama 2, Mixtral, and Grok.
#OpenAI releases seven Sora videos from select filmmakers, artists, ad agencies, and musicians; Fairly Trained CEO Ed Newton-Rex calls the move "artistwashing".
The AI Act is done. Here’s what will (and won’t) change
The companies with the most powerful AI models will face more onerous requirements, such as having to perform model evaluations and risk-assessments and mitigations, ensure cybersecurity protection, and report any incidents where the AI system failed.
I doubt it's coincidence that “GPT-5 is on the way!” news cropped up after some key #AI industry analysts praised Anthropic's Claude Opus as better than GPT-4. Large language models at this scale may be new, but tech vendor strategies are not.
Fairly Trained certifies KL3M, an #LLM claimed to be built without the permissionless use of copyrighted materials by legal tech consultancy startup 273 Ventures.
Here’s Proof You Can Train an #AI Model Without Slurping Copyrighted Content
YouTube adds new AI-generated content labeling tool
creators are required to disclose “altered or synthetic” content that seems realistic. That includes things like making a real person say or do something they didn’t; altering footage of real events and places; or showing a “realistic-looking scene” that didn’t actually happen