We need superknowledge before superintelligence
exa.aiThis is just blatant self-promotion. I'm not a fan of Google either, but running my last 5 searches through Exa I'm not even getting DuckDuckGo-quality responses. And you're advertising yourself in the corner, patting your own back in blogposts, ushering me towards a pricing page and telling me scary stories before I go.
Maybe we should stop super-washing things and sell products on their literal merit. Google isn't the premier search provider because they're "super" anything, they're successful because they're free and get out of your way.
Also, the current AI advancements result from walking away from knowledge-semantic-logic AI methodologies into more statistics-generative approaches. This is also a result of those methodologies not working properly because of the unsolvable problem of infinite knowledge = infinite space (a HD/compute unsolvable problem) // Infinite divisibility / synthesis of knowledge (also a physical/compute/HD unsolvable problem. Right?
I am having trouble deciding if he is arguing for a better api and a dystopian level of knowledge on individuals amd private businesses or just for a better api and thr examples are just mostly super dystopian.....
What does this company need an all-seeing eye for? Directory listings of every building and recursive search on the long-tail of typical search engine blind spots don't seem profitable enough on their own.
> "We’ll soon have near-AGI intelligences (GPT-5)"
Does anyone technical believe GPT-5 will be even remotely close to anything which has even a vague resemblance to AGI?
I'm technical and it wouldn't be surprising if GPT-5 will be able to code up small projects like an intern, perform complex actions across the web like an executive assistant, understand images and video better than most humans.
If you described this to someone 20 years ago, I think this would qualify as near-AGI intelligence.
It's totally possible that getting to true AGI -- where you could trust it with pretty much any complex task -- will require more breakthroughs.
GPT-4, GPT-3 and GPT-2 all coded up small projects, though. If GPT-5 is just another "it gets it right sometimes" dice-roll machine then it won't matter how many modalities it has. It will be DOA, and we'll all be pinning out hopes on GPT-6.
This shtick is running out of steam, and it's becoming increasingly clear that Sam Altman (or anyone at OpenAI for that matter) doesn't have what it takes to keep it going. The honeymoon phase is over, you can't promise capabilities that don't exist without proving that they're possible in the first place. Superintelligence is one of those things.
> If GPT-5 is just another "it gets it right sometimes" dice-roll machine then it won't matter how many modalities it has. It will be DOA
Every single thing and every single person can be seen as a "dice-roll machine"; when I give even a senior engineer a task to take on this sprint, there's a significant chance they won't complete it that sprint (and it might just have been infeasible at all). At the end, all work boils down to the project management triangle (On Time, On Spec, On Budget - choose two), and the better the AIs get, the more tradeoffs we can choose from and the more and more we'll gradually be using them, regardless of whether we'd able to ever trust them to complete a task perfectly 100% of the time.
If you want to be reductive, then sure. But even then, "superhuman intelligence" in this case literally means not being a dice-roll. I do not understand how you can pitch AGI without solving the hallucination problem, you might as well be selling people a random-number-generator that can do math with entropy.
Fascinating take, had never heard this term. Totally makes sense that intelligence/logic without access to peak information will keep spitting out subpar answers.
Am I the only one who had an issue with their example query of "give me all physics PhDs in NYC"? Are they really asking for a listing of all people living in NYC holding a PhD in physics? I can't shrug off the feeling that this is the sort of a request that The Terminator might make, just before filtering it further to those with the surname Connor. I for one feel that such queries should be somewhat difficult (at least beyond people who actively want to be found and put their credentials up on linkedin).
Key here is what is publicly available I think - agree though it's a fine line
clever take on reasoning vs. retrieval. the API is pretty clean, though some searches are still just ok.