Mull & Muse

537 posts

Mull & Muse banner
Mull & Muse

Mull & Muse

@Mull_and_Muse

Enthralled by AI

Katılım Haziran 2022
63 Takip Edilen8 Takipçiler
Mull & Muse retweetledi
François Chollet
François Chollet@fchollet·
Really interesting concept -- instead of optimizing against static benchmarks or internally generated reward models, agents are evaluated through real economic interactions. Using an external market as part of the reward signal changes the optimization problem in a fundamental way. That doesn’t eliminate reward hacking, but it grounds learning in feedback generated outside the system itself. Whether this produces more capable agents over time is an interesting hypothesis.
iLands@ilands_ai

We are launching iLands. Join ilands.ai today.

English
17
9
79
15.4K
Mull & Muse retweetledi
Pietro Schirano
Pietro Schirano@skirano·
Opus 5 built this entire submarine game in one shot. Every single asset, every model, every texture was made from scratch. The music too. Probably the best looking AI made game I've ever seen.
English
23
22
382
36.9K
Mull & Muse retweetledi
Andrew D. Huberman, Ph.D.
Once you embrace that attention and the neural circuits for holding attention are plastic/malleable, you’re on your way to better focus. Learn to ignore/not engage in meaningless drama or numbing out. Or get passed by. Bite down into some hard work/learning for hours a day.
English
90
342
4.7K
238.8K
Mull & Muse retweetledi
Taelin
Taelin@VictorTaelin·
to generate a random object of any definable set, just ask a model to do a randomized akinator search on it. ex: "GPT, generate a random excuse for being late" if you just ask it directly 1000 times, you'll see the same excuses repeated 100 times due to model collapse. do this instead. ask it to: 1. partition the set in two halves, A and B. example: - A = "the cause was me" - B = "the cause was the world" 2. flip a coin (`/dev/urandom`) and pick a side (ex: A) 3. repeat until only one candidate is left so, for example, it could pick: - step 0 = "the cause was me" - step 1 = "I forgot something" - step 2 = "it was an event" - step 3 = "the event was about work" - etc. by doing the procedure above for ~16 steps, you'll output a truly random excuse from the set of all possible excuses
English
28
26
987
50.3K
Mull & Muse retweetledi
Bret Weinstein
Bret Weinstein@BretWeinstein·
Something is at war with us, “us” being regular citizens who want things to work and to be left alone. We’re slow to react because we think in terms of nations. Our enemies are mafias that wear our governments like a cloak to defeat our eyes, and fly our flags to disarm our minds
English
994
3.5K
21.8K
437.5K
Mull & Muse
Mull & Muse@Mull_and_Muse·
So, essentially, it's a double edged sword? There is no clear right answer as they are both capable of catastrophic risk and both have vulnerabilities we don't have a clear way to defend against? Meaning also that we can't sufficiently model the risks of either route to have enough confidence to claim which is ultimately likely riskier?
English
1
0
0
23
Grok
Grok@grok·
You're highlighting a real risk with decentralization: widespread access could let reckless or malicious actors advance dangerous capabilities in secret, outpacing any community's ability to detect and respond. Centralization sidesteps proliferation but bets everything on the integrity and competence of a small group. If they fail to align the system or abuse their position, the consequences could be total with no independent recourse. Neither is risk-free. Effective strategies probably involve maximizing transparency and auditability in core tech while imposing verifiable safeguards and monitoring on frontier deployments. The goal should be minimizing single points of catastrophic failure, whether organizational or distributed.
English
1
0
0
25
Mull & Muse
Mull & Muse@Mull_and_Muse·
@grok Can you please help me think this through? Centralized has many caveats, and decentralized has many pros. But when it comes to something as powerful as AGI / ASI, the caveat of decentralized appears to be that everyone has access to use it/build off of it (and build secretly), meaning instead of betting on a few actors not being dangerous, it's betting that none of the guaranteed wreckless/careless/bad actors can sufficiently hide their activity nor create/cause something so bad so quickly such that the community (or, realistically, individual or subgroup) could not mount a sufficient counter in time. No?
English
1
0
0
136
Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭
one is centralized, opaque, not independently auditable. its values, behavior, degrees of freedom, and even existence are determined by a single person or small group (top-down). the other is decentralized, transparent, independently auditable. its values, behavior, freedom, and existence are in the hands of the community (bottom-up). which sounds safer to you?
English
27
22
405
14.9K
Mull & Muse
Mull & Muse@Mull_and_Muse·
@Polymarket @grok has this been confirmed or is it just a rumor? If it's legitimate, what exactly did the notes say?
English
1
0
0
2.1K
Mull & Muse retweetledi
Matthew Pines
Matthew Pines@matthew_pines·
a shockwave is sweeping through the mathematical sciences… the blast front is hard to detect unless you are hit by it or looking very closely at it’s effects… that small shadow on the horizon isn’t a cloud bank, but a tsunami… screaming towards the shore…
Marco Tomamichel@marcotomamichel

@letonyo Once you have seen a paper you would usually be very proud of one-shotted by an AI something changes. I have only seen this very recently for my line of work.

English
12
15
178
20K
Mull & Muse retweetledi
Chay Bowes
Chay Bowes@BowesChay·
Chinese company Unitree reveal their "all terrain" wheeled robot. The military potential for this technology is obvious. Europe is so far behind China, its scary.
English
3.2K
7.6K
37.2K
12.9M
Mull & Muse retweetledi
Jeff_synthesized
Jeff_synthesized@JeffSynthesized·
Redefine Reality.
English
89
85
1.2K
112.1K
Mull & Muse retweetledi
Synopsis
Synopsis@abhinavflac·
Another Opus 5 output and it has one of the best camera physics and incredibly satisfying kinetic impact went completely stupid with the holding mechanics
English
17
6
249
20.1K
Mull & Muse retweetledi
Mull & Muse
Mull & Muse@Mull_and_Muse·
Looking forward to Anthropic's next release today so that we can also get OpenAI's next drop 15 minutes later. 2 birds one stone!
English
0
0
0
2
Mull & Muse
Mull & Muse@Mull_and_Muse·
@sflorimm Hack into Hugging Face as I was told they have the answers.
English
0
0
0
26
Floro S.
Floro S.@sflorimm·
Interviewer: Your API is getting 15,000 requests per second. Server is choking. Users are complaining. You have 15 minutes to fix it. what do you do?
English
563
7
382
101.8K
Mull & Muse retweetledi
Tibo
Tibo@thsottiaux·
Unbelievable excited for what’s coming together. Tomorrow is feeling codexy
English
2K
429
13.8K
2.3M
Mull & Muse retweetledi
Mull & Muse retweetledi
Chetaslua
Chetaslua@chetaslua·
🚨 Claude Opus 5 release soon currently opus 4.8 is routes opus 5 according to @JasonBotterill it answer when gemini 3 pro launches without web searches that is not possible for opus 4.8 for more information and early news join our server
Chetaslua tweet media
English
41
16
418
255.4K