AI isn’t good enough • Hacker News Board

Re: AI isn’t good enough

#371

Post by arisAlexis » Mon, Aug 28, 2023, 2:06 PM UTC

bamboozled wrote: post #344

arisAlexis wrote: post #289

Earlier quoted context omitted.

I haven't written an article asserting my argument with certainty. He did. Big difference.

We’ll you have a opinion either way.

My opinion is that we don't know the limits. This is essentially no opinion because of no data. Stating a limit without data is an opinion thought and a wrong one.

Re: AI isn’t good enough

#372

Post by lynx23 » Wed, Aug 30, 2023, 4:49 AM UTC

pdimitar wrote: post #186

lynx23 wrote: post #179

Earlier quoted context omitted.

A toddler capable of walking is still miles away from the topic at hand, a LLM capable of processing relatively complex text. It is well know that humans have a particularily long training period compared to other animals. So I dont understand why you are bringing up the walking-training. It seems quite unrelated.

Because humans learn from much less samples sizes than "AI". The overly optimistic point of view on "AI" is over-represented in HN and that makes the discussion unbalanced and puts nearly everyone in it in a filter bubble. That's not OK.

Well, let me tell you a story. I have early beta access to GPT-4 with the ability to add pictures to the chat. This is rolled out as part of BeMyEyes, a volunteer-based app which lets volunteers have access to the camera for helping blind people. Now, BeMyAI lets me take a picture, send it to the AI, and have it describe the picture to me. Including OCR and text translation. And I can ask questions, interrogating the AI about the current picture. I can also add more pictures to the "chat", and the AI can see all of them, and even compare differences. I am 100% blind. After a few days of testing, I already know the AI is better then most humans I'd randomly ask to help me with pictures or scene descriptions. And it is more capable, as it can translate between every language I've encountered so far. If you ask me, the AI-will-kill-the-world crowd has a overly pessimistic view when it comes to AI, or they have been cought in the recent alarmism movement, or something. GPT-4 with picture integration is such a powerful feature, I'd say we haven't had anything equivant in terms of flexibility in the assistive technology market since, what, 30 years? This feature will liberate a few millions of blind people from the barriers put in fromt of them by society. And guess what, society didn't really solve them so far, OpenAI solved them as a side effect. But IT-activist still insist on seeing evil and demise everywhere.

Re: AI isn’t good enough

#373

Post by pdimitar » Wed, Aug 30, 2023, 1:41 PM UTC

lynx23 wrote: post #372

pdimitar wrote: post #186

Earlier quoted context omitted.

Because humans learn from much less samples sizes than "AI". The overly optimistic point of view on "AI" is over-represented in HN and that makes the discussion unbalanced and puts nearly everyone in it in a filter bubble. That's not OK.

Well, let me tell you a story. I have early beta access to GPT-4 with the ability to add pictures to the chat. This is rolled out as part of BeMyEyes, a volunteer-based app which lets volunteers have access to the camera for helping blind people. Now, BeMyAI lets me take a picture, send it to the AI, and have it describe the picture to me. Including OCR and text translation. And I can ask questions, interrogating the…

Well that's just super wholesome to read, thank you for sharing!

These and exactly these are the tech advancements that I imagined as a teenager would improve people's lives!

I just feel that 99% of the technology nowadays does not improve almost anyone's life. It gets easier and easier to be jaded and lose hope.

But your story is very reassuring and nice to read. Thanks for sharing!

Re: AI isn’t good enough

#374

Post by glenstein » Thu, Aug 31, 2023, 9:57 PM UTC

jimbokun wrote: post #267

tlarkworthy wrote: post #119

Earlier quoted context omitted.

AI has been around the corner since the 1950s, this is the historical evidence for the pessimistic stance against over optimistic predictions. LLMs are a huge stride forward, but AI does not progress like Moore's law. LLM have revealed a new wall. Combining multi agents is not working out as hoped.

And AI has been consistently successful since the 1950s, steadily achieving more and more that was once only the purview of human beings.

Right, I believe this is the more normal framing. I think Hubert Dreyfus spent his life constantly retelling the story of the 50s and 60s with a pessimistic narrative, and that is what made it stick, to the extent that it has. But it is a bizarre framing that, even if it once looked like a guiding light on how to set AI expectations, I think talking about the 50s and 60s over and over is not a helpful way of engaging with what has unfolded over the past 2-3 years.