Personality Types And Artificial Intelligence | Podcast 0470

9 comments

• 

  • Eric
  • June 1, 2023 at 4:25 pm

  For those who haven’t watched the movie “Her” – go watch it.

  The movie seems to me full of feeler types and constitutes a sort of crash course on Intuitive Feeling slices of life.

  Especially when the main character discovered his AI girlfriend was in “love” with over 600 others….. lmao there must be an ExFx metaphor in there somewhere

• 

  • Eric [INTP]
  • April 28, 2023 at 10:45 pm

  New development that’s occurred since March, but the activity is accelerating by the week-

  Facebook/Meta was developing their own LLM system called LLaMA, something like Large Language model for Meta AI or something, and they were sharing it with select researchers in a trial program. The software was fairly open but the critical piece, the “weights” files, were kept proprietary and shared only with program participants.

  Except someone leaked the weights, on BitTorrent, and the torrent link is in a pull request on Facebook’s github. Cat’s so far out of the bag right now as my coworker put it, this “leaker” is channelling Prometheus bringing fire to the people.

  Since then, a veritable Cambrian Explosion of DIY/community-fine-tuned/trained variations of the LLaMA weights have been released, and new ones are showing up daily at this point with varying levels of fine-tune.

  A critical development that’s showed up from the community is the invention of 4-bit quantized weights files, which allow larger neural networks – 13-billion, 30-billion, 65-billion and the like, close to being on-part with ChatGPT size – to run on commodity hardware, in some cases even CPUs, no expensive GPUs required.

  I have some old hardware – an HP server with ~2014-era CPUs – but I have had a blast playing with the Vicuna 13B models lately.

  The derivations are mostly named after the Llama theme somehow, e.g. Alpaca, Vicuna (genetic predecessor to the Alpaca), Red Pajama, etc.

  Folks are using spot-instance cloud resources of high-end servers with lots of GPUs to do the training/fine-tuning, and then releasing the “deltas” (need the original LLaMA weights to render them into usable shape) to keep above the legalities I guess. Problem is, so many community members have the original LLaMA weights you see “final” weights in a variety of formats (suitable for a variety of LLM model engines from GPU to CPU-based) posted on huggingface.co (an AI/ML community site ala Github) within hours.

  It’s a feeding frenzy. The hour of AI is well upon us, and the Cambrian Explosion is well underway.

  Few links-

  Old primer- https://www.reddit.com/r/LocalLLaMA/comments/11o6o3f/how_to_install_llama_8bit_and_4bit/

  llama.cpp (CPU-bound LLM execution engine)- https://github.com/ggerganov/llama.cpp
  This project is moving quick. I end up doing a “git pull” and recompiling the software several times a day.

  Salesforce-Codegen – https://huggingface.co/models?search=salesforce+codegen – a model for code generation, run by a fellow at Salesforce who seems to have been allowed to release these public. I’ve run this with another project, based on llama.cpp, called Turbopilot- https://github.com/ravenscroftj/turbopilot

  Turbopilot is compatible with the Github Copilot API and a plugin for Visual Studio Code called “Fauxpilot” exists to run it. It’s very slow on my server though but it’s promising.

  https://github.com/nsarrazin/serge – LLaMA.cpp Made Easy. Haven’t tried this yet but it sounds cool. Closer to “turn-key” AI running on your desktop computer w/o needing ChatGPT.

  Very latest Vicuna-derived models as of 4/28/2023- https://huggingface.co/TheBloke/stable-vicuna-13B-GGML – this is in GGML format for llama.cpp

  https://reddit.com/r/localllama/ is a good forum for keeping tabs on the latest developments.

  A description in this post mentions the different model data formats (GPTQ, GGML, HF, etc)- https://www.reddit.com/r/LocalLLaMA/comments/13279d6/comment/ji3ziq0/?utm_source=reddit&utm_medium=web2x&context=3

• 

  • Eric [INTP]
  • April 11, 2023 at 4:43 pm

  On the subject of what is ChatGPT interaction – a Question, or a Conversation, I say Will’s response was the most diplomatic but I err on the side of it being something you Question.

  Conversation, at least 2-way, requires 2-way feedback IMO, and ChatGPT is not learning from you in real-time. My understanding of its system (from a coworker who’s in charge of AI efforts at my employer) is that OpenAI employees are training the system separately in separate computer systems/processes, and releasing “updated training models” and we consumers are just running queries against the current “production” model.

  A true human being’s brain would be learning from the conversation in real-time, with 2 people providing bidirectional feedback. In this case, when you ask “followup” questions of the ChatGPT system in a particular thread, I think you’re only “appending to your original question” and it’s processing them as a whole (or using the prior section as bias for answering your next question). Create a new Chat thread and it’s like it forgot everything about the other threads. So this is really 1-way feedback – YOU are adapting, wrapping your mind around ITS answers, and adapting with followup questions, but it’s not adapting to you.

  Anyway that’s the most nitpicky INTP thing I could offer so I’ll stop while I’m still a nerd 8^D

• 

  • SAM
  • February 22, 2023 at 2:36 pm

  Hi, great episode! Just an opinion from one INTJ. But my “eye rolls” are mostly toward the excess of heeded warnings Chat GPT gives that have nothing to do with the original question. But I also see it as a necessary (evil) good put in place to prevent ignorance for those who may put their full faith in AI’s answers. Especially after the news of Microsoft’s Bing Bot going off the rails. Scary!

• 

  • Heather
  • January 29, 2023 at 12:55 pm

  So, after thinking about the comment I, myself, made above I would like to reform the idea a bit and since I cannot edit my original comment, I am making a new one. (As an ENTJ, when I get excited about something, I get a little bit silly, and have to mitigate my embarrassment with amusement and, sometimes, clarification.) In thinking about it further, it doesn’t make sense that there will be a physical counterrevolution unless along with AI in the future we create a sterile-ish, mechanistic society to go along with the AI (think George Jetson and Rosie the Robot.). It makes more sense that the shift will be mental, philosophical, emotional, etc. A renaissance of ancient knowledge and wisdom. The best part is that, at that point, it will seem “new” again. “We shall not cease from exploration, and the end of all our exploring will be to arrive where we started and know the place for the first time.” T.S. Elliot. The best part is, this sort of turning over has all happened before in different clothing. No matter our levels of achievement, nor the changes in the structure of our society, I do not think we will change what mankind is (not that anyone is necessarily making that claim, although some might be.). I have been playing around with the ChatGPT for the last couple of days and it is really fun! It’s nice to find a way to “play.” That’s what it feels like – play and discovery. Thank you for the podcast, and so many of the pod casts before. I have learned SO much from you guys and it is changing my life significantly. I’m super grateful.