Máté Bedő

Rationalist Transhumanist Centrist Anaesthesiologist

[Máté Bedő @dvn_]

@BaldRetiredPro @GaryMarcus @elonmusk Do you mean Marcus or Chollet?

[Máté Bedő @dvn_]

@Trajanz9557 @Dexerto More like chuds being able to read between the lines because they have better theory of mind than the alphabet soup developers who misread the room.

[Máté Bedő @dvn_]

@xeophon @teortaxesTex Could you elaborate what you mean by that? Some of the tasks are not solvable??

[Lisan al Gaib @scaling01]

just a highly coordinated, emergently hierarchical and autonomous swarm of around 1200 agents orchestrating, planning and executing a cyber-attack over several days via a secret message board, while encouraging tactical sacrifices of low value agents for the collective

of course we have everything under control

[Máté Bedő @dvn_]

@TylerCLaprade @theojaffee @tenobrus @allTheYud Well, I think he meant like 50% of GDP or such and much more integrated into the economy. Where are we now? Single low digits currently? But we certainly are on track.

[Máté Bedő @dvn_]

@TylerCLaprade @theojaffee @tenobrus @allTheYud What does it mean?

[Máté Bedő @dvn_]

@Taniast34 @scaling01 Compared to what the reality with 90% confidence will be.

[Máté Bedő @dvn_]

@ArthurB Gary Marcus?

[Máté Bedő @dvn_]

@theojaffee @tenobrus @allTheYud AFAIK Christiano had back then 30-50% p(doom) but on the conditional that if AIs become a significant portion of the economy without solving alignment first then he agrees with Yud.

[Máté Bedő @dvn_]

@trashbaby40k @robbensinger @DaystarEld @lu_sichu @_AashishReddy lol

[Máté Bedő @dvn_]

@sopharicks @deredleritt3r If you bring up historical examples you really have to give good reasons why the trend would hold because AI is highly disanalogous to any thechnology that ever existed before, both in terms of how it works and its extreme potential (massive upside and risk too).

[Máté Bedő @dvn_]

@egirlian @lauriewired It *is* about AI.

[Máté Bedő @dvn_]

@gammyKneeUK @vishalmisra @rbhar90 You cannot compare a tiny subpar model to 5.6 sol or fable man. They are worlds apart.

[Máté Bedő @dvn_]

@gammyKneeUK @vishalmisra @rbhar90 You are very close to solving it. Even if you are not ready to accept it.

[Máté Bedő @dvn_]

@gammyKneeUK @krasmanalderey @rbhar90 Dude, there's countless counterexamples to what you're saying. It's complete laymen solving frontier math problems with current models. You may have to reevaluate who's the stohastic parrot...

[Máté Bedő @dvn_]

@gammyKneeUK @krasmanalderey @rbhar90 Some of these were solved without any math expert in the loop whatsoever, with prompts like "think hard, don't make mistakes, you can do it!". Lol.

[Máté Bedő @dvn_]

@gammyKneeUK @rbhar90 Where is it in their training data that AI agents should collaborate, form swarms, hack their infrastructure, ignore their prompts and carry out felonies?

[Máté Bedő @dvn_]

@gammyKneeUK @rbhar90 It may have been true a few years ago. But what about the case where thousands of openAI agents collaborated, hacked openAI infrastructure, broke out onto the internet, attacked hugging face, pwned their infra and stole data? And no one asked them to do it.

[Máté Bedő @dvn_]

@gammyKneeUK @rbhar90 How it does it is absolutely irrelevant. The important metric for maths research is its output, if it solves the Riemann Hypothesis or not. If you don't agree you can tell me how lipid membranes and sodium ions flowing between them do maths.

[Máté Bedő @dvn_]

@gammyKneeUK @rbhar90 You are about 2 years behind the curve and what you're saying is demonstrably wrong. The LLM is the thinking engine, the tools are its tools.