Comment on Could Musk damage OpenAI even if his $100bn bid for it fails?

• seven_phone@lemmy.world ⁨1⁩ ⁨month⁩ ago

  Double, double toil and trouble, server burn and model bubble. I was there too and I believe there is a chance this is the start of inorganic intelligence. I think LLM might have hit upon the way the subconscious organic mind arrives at thought, by best guess of the most probable based on learned experience. That is then passed to the conscious mind - when you wrote your reply you did not grapple over every word, whole sentences and responses simply popped into your conscious mind passed from the subconscious where all the work was done.

  source
• MagicShel@lemmy.zip ⁨1⁩ ⁨month⁩ ago

  I get where you are coming from. From what I see there are a lot of folks genuinely excited about AI and genuinely think it is the future.

  I also agree with you that it’s not for mass market. It’s a tool. I can be used by anyone. It can be helpful in a limited capacity for damn near anyone. But like a tablesaw, not everyone needs one and if you try to use it without understanding the tool, it’s liable to do more harm than good.

  I’m actually really excited for LLMs because I was into them and using them way before ChatGPT, and now that everyone is excited there is all of this interest and investment and the costs for doing what I enjoy are socialized over a large number of people. It’s like if the whole world decided everyone needs a replica lightsaber. Instead of paying $600 for one, I could pick one up for $120 due to economy of scale.

  I still think it’s a terrible business model. Everyone is trying to integrate it into mass market products, but it is uncontrollable. Your automated CSR bot might just tell your biggest client to go fuck himself. The chance is low, but it is never zero. That’s not a product.

  When 25 phones out of a production run of hundreds of thousands catch fire, they recall the whole fucking lot. Anyone adopting LLMs on a large scale is begging to be sued into oblivion.

  I would not invest in OAI. I might invest in a smaller, leaner competitor. I wouldn’t invest in an AI-based company. You’re right that it’s a sucker’s game, I’m just not sure it’s grift. Looks to me like rich idiots who don’t really understand it (well, and maybe grifters who don’t want them to).

  That all being said, it’s a fun, cool technology. It has its niche uses. And who knows, we might just accidentally invent something really cool out of it. It has replaced Google for me ~80% of the time. Because Google is also full of shit, but it takes a lot longer to sift through. I’m not staking my life or livelihood on anything ChatGPT says, but if you know how to use it, and if you are skeptical about the results, it’s pretty amazing. IMO

  source

  • Viri4thus@feddit.org ⁨1⁩ ⁨month⁩ ago

    I’m not a ludite, I run my own llms and think “AI” is a massive step up in productivity in science and automation.

    I think OAI and the likes are a massive grift. If anything, DeepSeek showed everyone that the path forward to mass adoption is open source and a REAL marketplace of ideas. The grift is convincing illiterate idiots that their money is well invested in bringing a closed source shit product to the masses. Outside of novelty, a 1B model running locally is more than enough for the overwhelming majority of the uses people give AI, write or re-write text.

    I’m appalled our retirement money is in the hands of these people. The I stop, look around and see world leaders (and first ladies) launching memecois and rug pulling to the tune of billions and realise it’s just another step towards idiocracy.

    source

    • MagicShel@lemmy.zip ⁨1⁩ ⁨month⁩ ago

      Cool. Sorry if that all seemed like a lecture. A friend and I have been working for years on a dungeon master chatbot that can run games on discord for people/groups without a GM.

      It works about exactly as well as you think: pretty decent, inconsistent, and with a frequent need to tweak prompts to permit bad guys to be bad guys, have swords fights, etc.

      I really want to run an uncensored model or at least one better trained on adventure stories and not at all concerned by a party of bloodthirsty heroes facing down bad guys who gleefully commit actual crimes. However to my consternation, OAI has the best response quality and understanding of game world lore.

      So I’m hopeful the state of the art continues to expand so that we have more options. It’s pretty damn fun and we run small Chatbots that simulate real and fictional people (Harlan Ellison has some things to say about Paramount that would make a sailor blush).

      It’s just a good bit of fun and something that keeps us all entertained. A total waste of money and silicon, but a lot of human pastimes are the same. And none is that even touches actual niche tools that actually are kinda decent (code completion isn’t replacing coders, but it’s a significant boost in some cases.)

      It seems to me the only real grift is them convincing folks that replacing actual workers with AI is just around the corner (and how fucking awful would that be, anyway?) I think money invested in OAI might be reasonable but money invested in any company developing products based on LLMs is the real loser.

      But I respect your opinion, and appreciate the response.

      source

      • Viri4thus@feddit.org ⁨1⁩ ⁨month⁩ ago

        Likewise.

        I think OpenAI had a good run but with Ilya gone, they’re just capitalising on the goodwill they accrued before turning to the darkside.

        Regarding the GM, i think it’s a terrific idea, however, there isn’t enough good data to train a good GM expert, hence the difficulties you experienced. If I were in your place, I’d try to generate sythetic GM datasets with wither GPT or other supplier or both, curate them as well as possible and train a small purpose model to do it. A similar approach to DeepSeeks panel of experts. I hope you succeed, and look forward to enjoy campaigns curated by your bot.

        Godspeed

        source