this post was submitted on 19 Oct 2023
535 points (96.7% liked)
Technology
59288 readers
4364 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
ChatGPT is 10 months old, not even a whole year. And it was never fined tuned for story writing in the first place. A little bit premature to proclaim what AI can and can't do, don't you think?
ChatGPT isn't the entirety of AI, AI research has been going on much longer than ChatGPT has been around
Isn't chatGPT just a beefed up predictive text engine?
Yes. Honestly it's crazy how much people read into ChatGPT, when in practice it's effectively just a dice roller that depends in incredibly big dataset to guess what's the most likely word to come next.
There's been some research about this, the fact that people are assigning intelligence into things that ML does. Because it doesn't compute for us that something can appear to make sense without actually having any intelligence. To humans, the appearance of the intelligence is enough to assume intelligence - even if it's just a result of a complicated dice roller.
And that's exactly why we should be scarred. ChatGPT is just the popular tip of the AI iceberg, there is a whole lot of more stuff in the works across all kinds of domains. The underlying AI algorithms is what allows you to slap something like ChatGPT together in a few months.
AI has been being developed for 50 years and the best we can do so far is a dunning-kruger sim. Sure, who knows what it "can do" at some point, but I wouldn't hold my breath.
The recent deep learning AI efforts only started around 2012 with AlexNet. They were based on ideas that were around since the 1980s, but they had been previously abandoned as they just didn't produce any usable results with the hardware available at the time. Once programmable consumer GPUs came around that changed.
Most of the other AI research that has been happening since the 1950s was a dead end, as it relied on hand crafted feature detection, symbol logic and the like written by humans, which as the last 10 years have shown performs substantially worse than techniques that learn from the data directly without a human in the loop.
That's the beauty of it. Most of this AI stuff is quite simple on the software side of things, all the magic happens in the data, which also means that it can rapidly expand into all areas were you have data available for training.
You smug idiots are proud of yourself that you can find a hand with an additional finger in an AI image, completely overlooking that three years of AI image generation just made 50 years of computer graphics research obsolete. And even ChatGPT is already capable of holding more insightful conversations than you AI haters are capable of.
ChatGPT was released Nov 2022. Plain GPT1/2/3 neither had the chat interface nor the level of training data and fine tuning that ChatGPT/GPT-3.5 had and in turn were much less capable. They literally couldn't function in the way ChatGPT does. Even the original Google paper this is all based on only goes back to 2017.
Yeah, LLM won't ever improve, because technology improving has never happened before in history... The stupid in your argument hurts.
Beside GPT-4 can already handle 32768 tokens, that's enough for your average movie, even without any special tricks (of which there are plenty).