221–230 of 297 posts
Earlier quoted context omitted.
Isn't Chat-Bison-001 Palm 1? Edit: It seems I can't use my free credits on Vertex APIs... Not nice.
I don't think so because the CEO mentioned Bison as one of the PaLM 2 models in the Keynote. If I remember correctly.
May be a weird takeaway, but I did find it strange how much the whole report focussed on misgendering as a safety issue. I agree it’s important to get right, but it seems like one of hundreds of safety/alignment issues and that many others are de-emphasised or ignored.
May be a weird takeaway, but I did find it strange how much the whole report focussed on misgendering as a safety issue. I agree it’s important to get right, but it seems like one of hundreds of safety/alignment issues and that many others are de-emphasised or ignored.
That said, it's something that is more controllable across languages. All people, in all languages, have a roughly equal distribution of genders, but not race/religion, etc. Japanese language text will have similar gender distributions to English, but likely not equal distributions discussing race. That makes it a much better litmus test for multi-lingual bias.
Most of the misgendering discussion (2-3 paragraphs?) was in the translation section, which makes sense. A lot of the first classes in foundation courses learning a foreign language revolved around pronouns (which don't work the same in every language). Gender may be implied or absent in some. For example, to say "she is a doctor" in Italian, you might say "è un dottore", which has no pronoun (literally "is a doctor"). If you use google translate to make it English, "he" is added, assuming the gender. The potential for bias here is obvious, but consider that LLMs often deal with more context than a single sentence - if you're translating or writing story about a female doctor (where the gender is available contextually), you want all the use of pronouns to align where it makes sense. If a LLM didn't "understand" the pronoun in Italian, you might not recognize it, but in English, if the same person's gender was mixed across sentences, it'd be hard to read.
Earlier quoted context omitted.
What if it’s legal elsewhere ? Too bad?
Like, why does that matter? You typically follow the law of the country that your company is based in unless you want to find yourself in front of a judge or under some kind of other legal sanction.
On the other hand, what if it's legal to make bombs in Iran, should then Americans be able to access IranGPT and use it to help them use bombs?
May be a weird takeaway, but I did find it strange how much the whole report focussed on misgendering as a safety issue. I agree it’s important to get right, but it seems like one of hundreds of safety/alignment issues and that many others are de-emphasised or ignored.
An expert survey of 738 researchers who published in NeurIPS and ICML was done last year[3]. Their median estimation that AI will have an “extremely bad” long term outcome is 5%, and 48% of the researchers estimate the probability to be at least 10%. This is worryingly high considering the absolutely catastrophic consequences of the scenarios.
A minority of very vocal AI researchers (e.g. Yann LeCun) dismiss these risks entirely and claim that people read too much science fiction. But when you listen to their interviews it's very clear that they have no idea what they are talking about and never actually read any scientific literature on the subject.
The study of AI risks is a serious area of academic research that is worked on by labs from Stanford[4], Berkeley[5], Carnegie Melon University[6], Oxford[7], Cambridge[8], and many MANY other universities[9]. Not people who read too much science fiction.
————
[1] https://arxiv.org/pdf/2206.05862.pdf
[2] https://longtermrisk.org/files/Sotala-Gloor-Superintelligent...
[3] https://aiimpacts.org/2022-expert-survey-on-progress-in-ai/
[4] https://web.stanford.edu/~chadj/existentialrisk.pdf
[5] https://humancompatible.ai/about/
[6] https://www.cs.cmu.edu/~focal/
[7] https://www.fhi.ox.ac.uk/research/research-areas/#aisafety_t...
[9] https://futureoflife.org/about-us/our-people/ai-existential-...
Earlier quoted context omitted.
I assume this is just a PR/IR-driven project to stay the "Google is Dead" headlines hence the budget, especially considering an oversized chunk was spent on the scaling law, doesn't seem they were serious about building a GPT4-killer. I wasn't aware autoregressive LLMs were still considered an existential threat to Google. What's the threat supposed to be, ChatGPT is just going to keep eating Google search market sha…
I agree that if training data is what matters, it is likely that no one can compete with Google with Google Books, which scanned 25 million volumes (source: http://www.nytimes.com/2015/10/29/arts/international/google-... ), which is approximately all the books. DeepMind's RETRO paper https://arxiv.org/abs/2112.04426 mentions a dataset called MassiveText, which includes 20 million books of 3T tokens. So we know Google…
So how do we actually try out the PaLM 2? The links in their press release just link to their other press release, and if I google "PaLM API" it just gives me more press release, but I just couldn't find the actual document for their PaLM API. How do I actually google the "PaLM API" for a way to test "PaLM 2"?
Google's docs on the APIs are up: https://cloud.google.com/vertex-ai/docs/generative-ai/learn/... The pricing is also now listed but free during the trial period, although it's annoyingly priced by character: https://cloud.google.com/vertex-ai/pricing#generative_ai_mod... Assuming ChatGPT's tokens are the equivalent of 4 characters on average (a fair assumption), the pricing of PaLM's chat and embedding APIs are the…
Earlier quoted context omitted.
Btw I've arrived at a different interpretation of the "Open" in OpenAI. It's open in the sense that the generic LLM is exposed via an API, allowing companies to build anything they want on top. Companies like Google have been working on language models (and AI more broadly) for years but have hid the generic intelligence of their models, exposing it only via improvements to their products. OpenAI bucked this trend an…
OpenAI does not need people defending their scummy pivot.
Salvaging your business from that sort of tantrum by working with MS is called surviving.
Earlier quoted context omitted.
Yeah fast, but also kinda garbage last time I tried it. Does it even show sources now?
You know, it doesn't that I can see. I recall seeing it with sources in the demo at I/O this morning. It is smarter than the previous beta, but yes, it's still throws some wild pitches, and without sources. Still a work in progress.
Earlier quoted context omitted.
I don't think so because the CEO mentioned Bison as one of the PaLM 2 models in the Keynote. If I remember correctly.
But would be interested to know if that was not the case. They seemed to be saying that PaLM 2 was rolling out. Also the pages say its a preview. So why would they be previewing the old model still?
> Generative AI Studio, Model Garden, and PaLM 2 for Text and Chat are moving from trusted tester availability to preview, meaning everyone with a Google Cloud account has access.
> Codey, Imagen, Embeddings API for images, and RLHF are available in Vertex AI through our trusted tester program, and Chirp, PaLM 2, Embeddings API, and Generative AI Studio for text are available in preview in Vertex AI to everyone with a Google Cloud account.
It seems like you are right and general PaLM 2 is available. Fine-tuned code-generation model (Codey) is not publicly available yet.