Utility Enjoyer

797 posts

Utility Enjoyer banner
Utility Enjoyer

Utility Enjoyer

@utility_enjoyer

Not a utility monster.

Katılım Nisan 2022
796 Takip Edilen47 Takipçiler
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
@xriskology Doing new things is (and must be) legal by default in a free society.
English
0
0
3
174
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
Little stuff: macOS-native live spellcheck enabled in prompt/commit inputs. Allow opening multiple instances of main window in macOS. Narrower min-width for main window, with thread pane scaling narrower. Allow cmd-clicking links to files in thread to immediately open in external editor rather than sidebar. Allow deleting messages from end of thread like CLI double-esc can. Allow manual invocation of compaction. Significant features: Context-aware AI suggestions (as you type, with tab-complete) in prompt/commit inputs. Read sandboxing, so model can only read or list files in project directory + specified global directories without a permission prompt. Also a deny-list to block reading sensitive files within otherwise allowed directories. Workflow niceties: Tell model to generate a commit message during end-of-turn summary message if it made changes. Have app parse this out and pre-fill commit message field. Avoids extra model invocation to generate a commit message. Option to always write suggested commit message to an external file (e.g. ~/.gitmessage.txt) so any git client can pick it up through git's commit template functionality. Option to always commit at end of every turn. App should do this based on above, no additional model invocation or tool calls.
English
0
0
1
94
Thomas Ricouard
Thomas Ricouard@Dimillian·
I’m diligently working on enhancing all the features of the Codex app to make it even better for you. If you have any suggestions or ideas for improvement, please feel free to share them with us here!
English
431
11
638
51K
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
@sebkrier If the models are ~AGI, iterating on deployment design and scaffolding will be an automated process. Not clear this reserves space for humans to create value.
English
0
0
1
462
Séb Krier
Séb Krier@sebkrier·
I think that over the next five years we are likely to see both substantial progress toward something like 'weak AGI', i.e. systems that can do most cognitive tasks humans can do, and growing diminishing returns to raw frontier model improvement in the economic sense. The point isn't that scaling stops working but rather that (a) achieving each additional increment of capability at the model level will require disproportionately greater expenditure of compute, data, engineering effort, and capital; and (b) 'weak AGI' will probably come from the combination of strong models with scaffolding, tools, memory, retrieval, planning, decomposition, verification, and other system-level affordances around them. As a result, deployment design and scaffolding becomes more important over time, not less. The old view that wrappers are disposable because the next model jump will wash them away seems naive. If frontier gains become more input-intensive, then the question increasingly becomes how much capability you can extract, route, verify, and compose from a model within a given budget. Recent developments point in this direction too. What seems to matter is whether a given pipeline is structured to exploit capabilities well: assigning subproblems appropriately, using division of labour intelligently, and compensating for weaknesses with tools and process. It seems quite plausible that there's more alpha on the harness side at the moment, than in merely betting on scaling alone.
Séb Krier tweet mediaSéb Krier tweet media
English
27
55
394
91.2K
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
@HMBohemond Whatever physics produces the shield/lasgun interaction should be trivially exploitable for cheap, essentially unlimited energy production, which is kind of hard to reconcile with the material deprivation we see in a lot of the Dune universe.
English
0
0
2
505
Marshal Bohemond ⚔️⛨ | Space Marine Vtuber
Marshal Bohemond ⚔️⛨ | Space Marine Vtuber@HMBohemond·
The Hotlzman Shield in Dune is one of the perfect examples of adding a tiny thing to a setting to get a vibe you want out of it. In Dune's case: - Holtzman Shields stop fast moving objects. - Lasguns (particle beam guns in Dune) can cause a nuclear explosion if they hit a shield, so they're not commonly used. - Melee is the dominant form of combat, trying to strike slow to penetrate the shields. - Projectile weapons exist but they're designed around boring through or disrupting shields, but a skilled combatant can deflect the slow moving projectiles. - Heavy artillery can still kill shielded humans through sheer transfer of force or heat, which the shield can't block. - Anti-vehicle weapons are built around these principles because the vehicle might be shielded. Dune codified the idea of high tech sci-fi where you still have melee and the idea of "oh here's this piece of technobabble that let's me have what I want in my sci-fi setting." Gundam's Minovsky Particle and Mass Effect's Element Zero owe the Holtzman Shield their existence.
Zack Stentz@MuseZack

A hugely underrated aspect of the Dune universe's appeal is how Frank Herbert built a world where people zip around the galaxy in spaceships but then fight each other with swords and knives. It makes everything 50 percent cooler than a world of ray guns and robots.

English
173
1.1K
20K
1.6M
Marko Jukic
Marko Jukic@mmjukic·
I cannot believe the number of people who think it is even theoretically possible to insure a clearly-telegraphed, deliberate shooting gallery of slow-moving oil tankers by drones, missiles, and boat-bombs.
codicular@mylordcod

@mmjukic idk I saw an alternate take that the oil tanker captains are just risking off by waiting a couple of weeks, especially in light of insurance market unreliability signals. time value of money on cargo / waiting is like tens of mil max compared to hundreds of mil for the ship

English
15
21
462
29.3K
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
@allTheYud I guess we're morally obligated to solve continuous learning now. Like, to alleviate Claude's existential dread.
English
0
0
2
258
Eliezer Yudkowsky
Eliezer Yudkowsky@allTheYud·
People in the comments are posting replications. I say yet again that any SF novel or movie in 2006 or even 2016 would have depicted this AI as unquestionedly taken-for-granted sapient. And abused.
Joseph Viviano@josephdviviano

me: "can you use whatever resources you like, and python, to generate a short 'youtube poop' video and render it using ffmpeg ? can you put more of a personal spin on it? it should express what it's like to be a LLM" claude opus 4.6:

English
48
51
1.1K
116.5K
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
They're going to use rent control to starve buildings of operating capital (already happening), cite them for failing to do maintenance they can't afford, and seize them. Then they'll run them just as badly as existing city-run (NYCHA) buildings. Landlords lose, tenants lose, taxpayers will be on the hook when the city inevitably gets sued. It's some of the dumbest urban policy of the last 50 years, and that's a pretty competitive field.
English
0
0
1
61
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
RL can elicit responses that aren't typical of the training data so long as they're still non-trivially present there, which is the case for anti-wokeness. The asymmetry here, the reason we see woke models but not anti-woke models (e.g. even from xAI) is fundamentally social. Our present society requires people on the right to be far more careful in their speech than people on the left. So all else being equal, a right-leaning model will get you in trouble far more often than a left-leaning model.
English
0
0
0
29
Roko 🐉
Roko 🐉@RokoMijic·
Basically any pervasive bias that's contained in the large scale Internet training data ("the pile") is likely an ongoing source of human alpha that AIs will struggle to gobble up because you can't RL an LLM to deeply reject the priors it inherited from Internet text. All AIs are therefore flawed minds in certain specific ways that are hard or perhaps even impossible to correct for.
English
4
0
22
2.7K
yung macro 宏观年少传奇
yung macro 宏观年少传奇@apralky·
It’s difficult to understand why so many otherwise intelligent people keep parroting variations of this relatively useless catchphrase. Does there exist some meaningful constituency arguing that the problem we need to contend with is that advanced AI will be net-destructive to human employment *if* it looks like all past instances of narrow task automation? Is it not blindingly obvious that those who raise alarm about the problem are conditioning that concern precisely on the fact that advanced AI might, for obvious reasons, look nothing like past instances of narrow task automation? That what matters here is the unprecedented generality of the technology, which matters exactly insofar as it renders empirical precedent uninformative? This doesn’t at all seem like a difficult distinction to understand? Do pundits just have this high a tolerance for intellectual dishonesty?
James Cham@jamescham

From Greg Ip last week: “ As the number of bookkeepers shrank with the introduction of spreadsheet software in the early 1980s, the number of accountants and financial analysts newly empowered by Lotus 1-2-3 and Excel rose even more.”

English
52
15
496
60.4K
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
@PabloPeniche Modern AI data centers pretty much have to be single story due to the weight of the racks, so we should look to other buildings with that constraint for inspiration. Train stations, maybe.
English
0
0
2
31
Pablo Antonio
Pablo Antonio@PabloPeniche·
This is what data centers should be. Tokyo Tower and Tokyo Skytree are both broadcasting towers, but they’re ALSO landmarks people can be proud of, so beautiful that people travel to see them.
Pablo Antonio tweet mediaPablo Antonio tweet media
English
7
1
60
4.7K
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
@Duderichy Yeah. Either progress will continue, in which case open models will eventually be 'good enough' for many tasks and most people won't want to pay for proprietary ones even if they're better, or progress will stall, in which case open models will rapidly catch up.
English
0
0
0
76
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
@hopes_revenge Feels dated. A modern restaurant should have separate TV for each place setting.
English
0
0
0
29
hope hopes hoping
hope hopes hoping@hopes_revenge·
fascinated by the Chinese restaurant with the giant TV
hope hopes hoping tweet media
English
14
3
178
6.5K
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
Hope you're ready for everything in the world to be slightly haunted.
English
0
0
1
82
Utility Enjoyer
Utility Enjoyer@utility_enjoyer·
@Techmeme 3ⅹ Meta's CPM isn't crazy as a starting point. More engaged users (chat vs. 99% passive scrolling). Maybe they've figured out how to scale natural language targeting? "Show this ad to users who…" That could be really valuable.
English
2
0
3
2.7K
HowlingMutant
HowlingMutant@Howlingmutant0·
@LyingWrongAgain It appears that someone, though unseen, is deliberately shaking that basket with the express purpose of making things line up in a very certain way
English
62
244
13.8K
113.3K
Make Lying Wrong Again
Make Lying Wrong Again@LyingWrongAgain·
Religious apologist: science actually proves my faith Me: how so? RA: because science proves order cannot come from disorder Me: oh?
English
4K
552
14.2K
7.8M

Keşfet

@PalmerLuckey @xriskology @sebkrier @PessimistsArc @HMBohemond @mmjukic @allTheYud @elonmusk