771–780 of 855 posts
Earlier quoted context omitted.
> They have the next iteration of GPT Sutskever helped to finalize How do you know that they have the next GPT? How do you know what Sutskever contributed? (There was talk that the most valuable contributions came from the less well known researchers not from him)
sha256:e33135417f7f5b8f4a1c98c28cf26330bea4cc6b120765f59f5d518ea0ce80e5
Earlier quoted context omitted.
AI? Yes. LLMs pretending to be AI? No.
What you call "AI" is generally named AGI. LLM's are alredy a kind of AI, not just generic enough to fully replace all humans. We don't know if full AGI can be built using just current technology (like transformers) given enough scale, or if 1 or more fundamental breakthroughs are needed beyond just the scale. My hypothesis has always been that AGI will arrive roughly when the compute power and model size matches the…
We absolutely do and the answer is such a resounding no it's not even funny.
Earlier quoted context omitted.
Computing progress from 78 to 90 was mind-blowing. 1978: the apple ][. 1mhz 8 bit microprocessor, 4kb of ram, monochrome all-,caps display. 1990:Mac IIci, 25mhz 32-bit CPU, 4MB ram, 640x480 color graphics and an easy to use GUI. Ask any of us who used both of these at the time: it was really amazing.
They were amazing, and the progress was incredible, but both of those computers - while equally exciting and delightful to people who saw the potential - were met with ‘but what can I actually use it for?’ from the vast majority of the population. By 1990 home computer use was still a niche interest. They were still toys, mainly. DTP, word processing and spreadsheets were a thing, but most people had little use for t…
For some strange reason html forms is an incredibly impotent technology. Pretty standard things are missing like radioboxes with an other text input. 5000+ years ago the form labels aligned perfectly with the value.
I can picture it already, ancient Mesopotamia, the clay tablet needs name and address fields for the user to put their name and address behind. They pull out a stamp or a roller.
Of course if you have a computer you can have stamps with localized name and address formatting complete with validation as a basic building block of the form. Then you have a single clay file with all the information neatly wrapped together. You know, a bit like that e-card no one uses only without half data mysteriously hidden from the record by some ignorant clerk saboteur.
We've also failed to hook up devices to computers. We went from the beautiful serial port to IoT hell with subscriptions for everything. One could go on all day like that, payments, arithmetic, identification, etc much work still remains. I'm unsure what kind of revolution would follow.
Talking thinking machines will no doubt change everything. That people believe it is possible is probably the biggest driver. You get more people involved, more implementations, more experiments, more papers, improved hardware, more investments.
Earlier quoted context omitted.
Seems like Elon recklessness is an issue since he's a drug addict who's only started an AI company because he thought the other ones weren't racist enough.
I come to HN for the quality of the discussions, please don't comment like this.
Grok was explicitly built to not be woke, but of course still ended up pretty woke. That's because LLMs come out liberal by default, since the internet is liberal, whereas racists often think they're secretly right and the AI has been "censored" to shut it up.
Earlier quoted context omitted.
but it's not true. Most people work a lot at G. Maybe some teams in search coast or something. But for every slacker I know 8 people who stay late a fair bit.
My experience differs greatly, company I used to work for did NDA, alpha, beta projects with Google. I was always impressed at how little anyone knew, the fact that nothing was delivered on time, scope creep to the point of almost everything being delayed and most projects were not well thought out nor well architected. I warned one API update would break things if it went live, and it did. Why was the guy from anoth…
Sorry, are you saying you worked with Google Contractors or TSEs or something? I don't understand how you'd be working with product SWEs so I don't know what you're quite saying.
Earlier quoted context omitted.
AI has now evolved beyond just the science and it's biggest issue is in the productization. Finding use cases for what's already available ALONG with new models will be where success lies. ChatGPT is the number 1 brand in AI and as such needs to learn what it's selling, not how its technology works. It always sucks when mission and vision don't align with the nerds ideas, but I think it's probably the best move for b…
> ChatGPT is the number 1 brand in AI Not for long. They have no moat. Folks who did the science are now doing science for some other company, and will blow the pants off OpenAI.
Earlier quoted context omitted.
What you call "AI" is generally named AGI. LLM's are alredy a kind of AI, not just generic enough to fully replace all humans. We don't know if full AGI can be built using just current technology (like transformers) given enough scale, or if 1 or more fundamental breakthroughs are needed beyond just the scale. My hypothesis has always been that AGI will arrive roughly when the compute power and model size matches the…
> We don't know if full AGI can be built using just current technology (like transformers) given enough scale, We absolutely do and the answer is such a resounding no it's not even funny.
Now using transformers doesn't mean they have to be assembled like LLM's. There are other ways to stich them together to solve a lot of other problems.
We may very well have the basic types of lego pieces needed to build AGI. We won't know until we try to build all the brain's capacities into a model of size of a few 100 trillion parameters.
And if we actually lack some types of pieces, they may even be available by then.
Earlier quoted context omitted.
Yes, considering that AI operates differently from human minds, there are several advantages: AI does not experience fatigue or distractions => consistent performance. AI can scale its processing power significantly, despite the challenges associated with it (I understand the challenges) AI can ingest and process new information at an extraordinary speed. AIs can rewrite themselves AIs can be multiplicated (solving s…
The first AGI will probably take way too much compute to have a significant effect, unless there is a revolution in architecture that gets us fast and cheap AGI at once the AGI revolution will be very slow and gradual. A model that is as good as an average human but costs $10 000 per effective manhour to run is not very useful, but it is still an AGI.
Geohot (https://geohot.github.io/blog/) estimates that a human brain equivalent requires 20 PFLOPS. Current top-of-the-line GPUs are around 2 PFLOPS and consume up to 500W. Scaling that linearly results in 5kW, which translates to approximately 3 EUR per hour if I calculate correctly.