Join the conversation

Join the community of Machine Learners and AI enthusiasts.

Sign Up
appvoid 
posted an update 9 days ago
Post
3475
suppose someone is working on a reasoning model, which ends up unlocking achievements that lead to agi, should it be open source?

keep in mind everybody will have access to it: scientists, governments, terrorists, average people, etc...

This question is subject to significant debate.

·

i know right?

In my pov, it should be open, if I can achieve AGI, someday someone will too. So theres no need to slow things down like eu. Just let things happen, accelerate and decentralize.

·

Also, I do not think someone will achive AGI as we dont know what AGI is. I think we will just do incremental perf insreases, not an "unlock" that creates AGI.

You are a victim of marketing.

"This system is so advanced it could destroy the fabric of society, create bioweapons, etc. But I guess I could give it to you for... $20?"

Sam was also "worried" about GPT-2. He's a salesman.

·

obviously gpt 2 was the kickstarter for openai but they didn't actually know the power of gpt4 when they created gpt 1

same thing could happen with reasoning, it might or might not have bigger implications, who knows

If AGI ever does happen, it will be open-source, eventually. There is no One True AGI. Literally some other lab could just make their own AGI afterwards especially because they'll have AGI to help them do it.

It's a weird question to ask. Like if AGI is inevitable, open AGI is also inevitable.

Sorry for the cynical sounding answer, but: If we did have open source AGI, tell me how it would in practice realistically change the power dynamic for better or worse?

The info to make jets, nukes, etc exists in the public domain, but 99% of threat actors structurally lack standing to build or leverage either. Same here with AGI and AI tech in general...

Suppose we did open source AGI:

  • The government, financial institutions, criminals and terrorists in the developing world, "entrepreneurs" with money behind them, real entrepreneurs starting with nothing, and everybody else 'alike' "will have access" to open source AGI.
  • The government, financial institutions, family offices, ... will have a team of quants, AI scientists, and engineers running it to their agenda with an unlimited budget for the hardware to run it on or unlimited inference credits.
  • Criminals, terrorists, and low income altruistic entrepreneurs alike will get their $0.10 of inference credits that will let them work for half a day before they hit a paywall.
  • Everybody else: Middle class "power users" will get their "sort of unlimited" scaled down version or a few prompts a month at the full sized model when the scaled down version isn't working.
  • ... And the last 2 categories will be fighting to outdo those in the first category. There will be 0.1% with a combination of whit, connections, and luck that will beat the dealer at his own table and will be a few people rise to power or pull off something that has a major impact, good or bad. For the other 99.9%, things will remain as they are.

What would open source AGI change then in terms of it posing a threat? How would a criminal or terrorist having access to a copy give them the upper hand over the rest of us?

·

100%! We had FOSS and corporations swallow it hole via famous $MS EmbraceExpandExtinguish, we also had Carnot cycle in open domain but no unemployed textile worker had the resources to build it's own mill with steam engine... ;)
We are hating the machines taking our jobs instead of hating the society making this a bad thing...