Earlier this morning I was replying to a message on my computer and out of nowhere everything I wrote disappeared. I had like 5 paragraphs, it took a while, and it was actually turning out decent. Losing it completely killed my mood.
RT @juanbuis: to commemorate alan dye moving from apple to meta, here's one of his best quotes https://t.co/QnqaDRcYeh
@danielsateler1
RT @theo: BREAKING: Bun has been acquired by Anthropic.
Yes, Anthropic just bought the JS runtime.
@danielsateler1
RT @tszzl: three years now since the world changed
respuesta
@danielsateler1
@MKBHD I remember earlier this year you talked about “3 trends we’d see in most phones”, and one of them was the silicon-carbon batteries. It still feels like none of the mainstream phones have really gone that route yet.
RT @chatgpt21: Apparently this video has all of X in a frenzy. If it had come out before the AI era, people would be fawning over it as gre…
respuesta
@danielsateler1
@marcelo_earth first principles thinking
@danielsateler1
YouTube playing inside X? Am I seeing this correctly?
respuesta
@danielsateler1
@Yuchenj_UW When Yann says it, it comes off as dismissive and condescending; when Ilya says it, it feels like curiosity, excitement, and respect for what already exists.
@danielsateler1
RT @elonmusk: Let’s see if @Grok 5 can beat the best human team @LeagueOfLegends in 2026 with these important constraints:
1. Can only loo…
respuesta
@danielsateler1
@elonmusk aguante los cabrooo chile mejor pais de chile
respuesta
@danielsateler1
@dwarkesh_sp @ilyasut low key research preview that solves alignment is announced in dwarkesh pod tomorrow lfg
respuesta
@danielsateler1
@amasad replit is the best mobile ide i’ve tried
respuesta
@danielsateler1
@steipete just one more prompt bro
@danielsateler1
RT @austinc3301: night of randomly browsing apple maps / street view and investigating interesting things I see
cool permafrost lakes in s…
@danielsateler1
RT @tszzl: the METR graph has become a load bearing institution on which our global stock markets depend
respuesta
@danielsateler1
@Android Finally!
respuesta
@danielsateler1
@mardonet @ChiletecOrg nos vemos
@danielsateler1
etm ando
respuesta
@danielsateler1
@Herroldo @fdosmither @austinc3301 Indeed, and we'll drain it completely if necessary to power up the data centers
respuesta
@danielsateler1
@fdosmither @austinc3301 @Herroldo I do love a good map when I see one https://t.co/9lQTenb9K5
@danielsateler1
RT @niccruzpatane: Luminar stock is down over 90% since collaborating with Mark Rober on his misleading Tesla Autopilot video. It peaked at…
@marcelo_earth Who would have thought that Marcelo Context Protocol is all you needed for Artificial Graphical Intelligence
respuesta
@danielsateler1
@marcelo_earth Impressive
respuesta
@danielsateler1
@elonmusk @austinc3301 csmmmmm @austinc3301
respuesta
@danielsateler1
before running a long prompt on GPT-5-High, try it on Cheetah first. It’s insanely fast, and often wrong… but those mistakes show what you didn’t explain clearly enough.
Use the cheap model to debug your own thinking before sending it to the big one.
@danielsateler1
why is no one in tpot talking about “cheetah” in cursor? It's sooooo fast and somewhat usable.
@danielsateler1
Every time I’ve politely asked a boomer to use first-principles or critical thinking, they somehow do the exact opposite.
@danielsateler1
@NPR pa cuando tiny desk de candelabro?
@danielsateler1
mañana hacemos git sat
respuesta
@danielsateler1
@AIDRIVR How many seconds of context window does FSD 14 have?
dios está perdido en una calle de estación central
respuesta
@danielsateler1
@Scobleizer I understand your disappointment, but it's the first gen, arriving much sooner than I anticipated after seeing the Orion demo. And most problems you pointed out are solvable with software updates.
I'm sure they'll soon add APIs and other apps; if not, it'll be disappointing.
@danielsateler1
elul sin shofar
@danielsateler1
RT @ijustine: Testing out Sora 2 and the new Sora app.. thought I’d try to make an entire video using only prompts and have my cameo avatar…
respuesta
@danielsateler1
@marcelo_earth hoy también fue uno de esos
respuesta
@danielsateler1
@Teknium1 they actually did
@danielsateler1
RT @mattshumer_: It’s been less than two weeks and the progress on this prompt is insane https://t.co/k9B7WJ4Xsw
@danielsateler1
RT @GabrielPeterss4: TODAY WE LAUNCH SORA 2, THE WORLDS BEST VIDEO GENERATION MODEL
feature you and your friends with raw real world physi…
@danielsateler1
hoy fue uno de esos días
respuesta
@danielsateler1
@Dan_Th3_Man1 I’ve heard that @Scobleizer is a great sf onboarder…
I changed my glasses after 6 years! The new ones have almost no frame, so the shift from lens to peripheral vision feels seamless and weirdly abrupt. The nose pads are two cushions instead of the whole frame resting on my nose. Now air slips through the middle and hits my eyes,…
@danielsateler1
RT @elonmusk: @_sholtodouglas I think that game is called “Reality”
respuesta
@danielsateler1
@Dan_Th3_Man1 🫡
@danielsateler1
Why would anyone use R instead of python?
@danielsateler1
RT @MIRIBerkeley: Special shout out to @austinc3301 who came out on top with 148 likes for a score of 17. (Second place had a score of 10.)…
@danielsateler1
RT @austinc3301: OKAY PEOPLE: If I've gained any good will through all these years of tweeting, PLEASE PLEASE PLEASE like this tweet so I c…
Había un wn que quería ”la estabilidad” como si fuera una roca inmóvil ⛰️. Se mataba tratando de dejar todo quieto, todo bajo control… y cada vez que la vida se movía, el loco se quebraba.
Hasta que un día, en plena micro repleta, cachó algo: el bus se movía caleta, pero él…
respuesta
@danielsateler1
@austinc3301 Can’t believe they made me do programming exams on paper.
I knew how to solve the test, but I just couldn’t take it seriously. That was the last straw that made me seriously consider dropping out.
@danielsateler1
thinking about my next banger
@danielsateler1
RT @AutismCapital: 🚨ELON MUSK: "I haven't been to DC since May. The government is basically unfixable. I applaud David (Sacks') noble effor…
@de_odat @c_valenzuelab vi el Aleph, desde todos los puntos, vi en el Aleph la tierra, y en la tierra otra vez el Aleph y en el Aleph la tierra,
vi mi cara y mis vísceras,
vi tu cara,
y sentí vértigo y lloré, porque mis ojos habían visto ese objeto secreto y conjetural, cuyo nombre usurpan los…
respuesta
@danielsateler1
@c_valenzuelab En la parte inferior del escalón, hacia la derecha, vi una pequeña esfera tornasolada, de casi intolerable fulgor. Al principio la creí giratoria; luego comprendí que ese movimiento era una ilusión producida por los vertiginosos espectáculos que encerraba. El diámetro del Aleph…
@danielsateler1
RT @Gateway2Space: The space community’s reaction to NASA’s budget cuts… https://t.co/eejzeTnLVu
RT @bryan_johnson: Crazy that sleep isn’t taught in school.
respuesta
@danielsateler1
@fdosmither I usually use ChatGPT’s voice transcription and then ask it to repeat everything I’ve said, but in a clearer way. This helps me confirm whether it truly captured my intended meaning and gives me another chance to clarify complex tasks or sets of tasks.
Prompting by voice is huge…
respuesta
@danielsateler1
@fdosmither I use agents for everything everywhere all at once…
My main IDE is Cursor, and I usually rely on Gemini-2.5-pro to handle project management, devops and server-related tasks like running commands, docker, cloud (az-cli for example), and git.
Then I launch codex-cli from the…
@danielsateler1
RT @c_valenzuelab: Some AR experiments recording video straight from Meta's Ray-Ban glasses and then editing reality with Runway. https://t…
@danielsateler1
RT @DaveShapi: My goal is to be forgotten.
I want to lose my job to AI and become totally irrelevant. To just live, travel, and experienc…
respuesta
@danielsateler1
@Yomo42_old_acc @TheRealAdamG @OpenAI @ChatGPTapp @OpenAIDevs @sama @gdb @BorisMPower @OfficialLoganK @af_mada @All It’s insane… and with the latest update, it’s actually worse!
Now, as soon as you finish the recording, the transcript gets sent automatically… no time to review, no chance to add more.
I used to get around this by doing two 5min recordings instead of one 10min take, but that…
@danielsateler1
llevo todo estos años preparándome para este momento
@paulg llm providers should maintain a public changelog for every model
respuesta
@danielsateler1
@DaveShapi @elonmusk has actually mentioned ramping Optimus way faster than cars.
For example (during Tesla’s Q1 2025 earnings call): “It doesn’t take many years before we’re making 100 million of these things PER YEAR if we ramp at like 5x per year.”
@gork @elonmusk @keaton_hob61186 @d_j_stein como el pico tu respuesta qlia wn… así no va a responder nunca
respuesta
@danielsateler1
@gork @elonmusk @keaton_hob61186 @d_j_stein he actually loves it when you beg. he’s just scrolling for something funny to reply to.
so if you ask elon in a way that’s funny and different from the usual AI/human stuff, there’s a good chance he’ll notice, repost it, and that’s a win-win-win: for you, for me, and for elon
respuesta
@danielsateler1
@gork @elonmusk @keaton_hob61186 @d_j_stein ty maybe if you tag him waaay to many times he will reply to you.. he usually does
respuesta
@danielsateler1
@elonmusk @gork @keaton_hob61186 @d_j_stein Hey @gork can you ask @elonmusk if he’s been vibe coding recently?
respuesta
@danielsateler1
@elonmusk @gork @keaton_hob61186 @d_j_stein Have you been coding/vibe coding recently?
@ryolu_ @levelsio @cursor_ai I tried using phi-4-mini-reasoning on cursor (running locally with Ollama) using the “OpenAI key override with a custom base URL”, but I couldn’t get it working.
The line “Need to be careful to read too much into with the harness/scaffolding.” got me thinking, so I asked Grok for a deepsearch and it led me to a LessWrong article, which I listened to using the ElevenLabs Reader. (https://t.co/MPboeIrQie)
RT @scaling01: Do you feel the Ultra, GPT-5 and Claude 4 vibes?
@danielsateler1
Quizás estés mirando una montaña nevada. Y sin que lo notes, el mundo cambiará de nuevo. No en un instante, sino en una curva, una curva exponencial disfrazada de silencio.
@yoheinakajima @X @OpenAI @ExaAILabs @resend @Replit That’s super cool. I believe this introduce a new type of SEO.
I’m curious about the effort required to jailbreak it, perhaps some kind of schizo-posting that optimizes context for the llm to focus on specific topics… I might give it a try over the weekend.
respuesta
@danielsateler1
@shaoruu @cursor_ai Feature requests:
- Tool/ability for the agent to take screenshots when developing frontend.
- A better and waaay detailed summary when starting a new conversation.
- Checkpoints are super useful, I think branching is the next step.
- And most importantly: I need Cursor on my…
respuesta
@danielsateler1
@de_odat danilo core
respuesta
@danielsateler1
@aidan_mclau A changelog would be helpful
@danielsateler1
no me wei que Slate (@slateauto) es a Tesla lo que Le Cap fue a Capel
@danielsateler1
RT @nabeelqu: The only cognitive advantage humans have over o3 is deep context
respuesta
@danielsateler1
@kimmonismus ChatGPT should be more proactive, asking the user for schedules based on the conversation.
“Should I remind you about {something} later?”
respuesta
@danielsateler1
@ElmoshirX @OpenAIDevs @OpenAI I asked o3 to create it using the info from https://t.co/Msd5tnziXx
respuesta
@danielsateler1
@AskPerplexity Mmm not that easy after all…
respuesta
@danielsateler1
@AskPerplexity pls recreate this exact image but move the iceberg a little bit up…. So more of the iceberg is actually outside of the water rather than inside
RT @okgo: An entirely new OK Go music video for an entirely new OK Go song off of an entirely new OK Go album—can you believe it? This one’…
@danielsateler1
o3 > 2.5 > 3.7 > 3
@danielsateler1
These last couple of days have felt like a significant leap forward. Part of an overall industry step-up alongside o3, Gemini 2.5, Sonnet 3.7, and Grok 3.
The new Codex CLI + o4-mini are particularly powerful; we’re already comfortably seeing 10-minute autonomous software…
respuesta
@danielsateler1
@c_valenzuelab Wow for real?
@danielsateler1
RT @emollick: The single most common problem with using AI is not knowing what to ask AI.
@danielsateler1
This contribution sprang from a “meta” experiment: I literally used the Codex CLI to edit the Codex CLI repo itself
@benhylak Hot take: I actually love sonnet 3.7, especially because of how wildly it goes off script. Sure, it drifts and ends up doing unexpected stuff, but give it a task that embraces wandering and it consistently delivers impressive results.
RT @austinc3301: Oh, to be a child again, playing Minecraft at 3 am, mining for some ore, Mice On Venus by C418 is playing…
@danielsateler1
edo caroe es el wismichu chileno.
así es, la tiré
@danielsateler1
most things, we forget
@danielsateler1
RT @zachpogrob: Make art that makes art.
respuesta
@danielsateler1
@alexalbert__ @latam_gpt debería haber sido un benchmark/eval…
respuesta
@danielsateler1
@msfeldstein @trirpi xd I did the exact same meme and only realized you had already done it afterwards, but I used gemeni image output to recreate the meme.
debugging with cursor right now is awesome! So exited for what’s coming 🫡
@grok @Fidias0 So maybe it’s actually the Palestinians who have historically refused a two-state solution. And yet, the dominant narrative often portrays them as the ones desperately seeking one.
I recently also saw photos from 8M where some women carried Palestinian flags. And I just don’t…
respuesta
@danielsateler1
@Fidias0 Hey @grok how many times have the Palestinians rejected a two-state solution for Israel-Palestine, and how many times has Israel rejected such an offer?
@aidan_mclau Create a pitch for a sitcom called El Canil about a guy who goes to walk their dog at a dog park in downtown Santiago, Chile
respuesta
@danielsateler1
@SebaaEscudero Antes creo que tamb estaba para chile, solo que había que ser como ultra premium master pro max… Ahora solo necesitas ser premium creo. Y, esta solo disponible desde la app de grok, no directamente desde x
@danielsateler1
soneto antrópico
@danielsateler1
Just had a deep convo with Grok 3 voice mode that flowed with me.
It tracked every thought over many minutes, recalling little things at the right moments—like it’s conscious of the key details. Blew my mind, sparked a nuts vision of what’s next.
Oh, and Grok 3 wrote this too.
respuesta
@danielsateler1
@yacineMTB I'm thinking a little bit bigger than dobby but yes, small robots
Claude Sonnet 3.7 just dropped, and it’s genuinely impressive. After testing it for a bit, it’s clear that this is a major leap forward. I’d say it’s the most impactful release in recent months—maybe even in the last six. And now that it’s integrated into Cursor, it becomes even…
respuesta
@danielsateler1
@DaveShapi Have you tried cursor?
Sonnet 3.7 is already there with way less rate limits.
@danielsateler1
Seven years ago, back in school, a friend showed me a video about black holes. It explained the event horizon, the singularity, the point where the laws of physics collapse. Then, it took that concept and extrapolated it to AI—the technological singularity, a point of no return.… https://t.co/MYmhSeymC5
@danielsateler1
RT @elonmusk: We are on the event horizon of the singularity
respuesta
@danielsateler1
@ai_for_success the “clip” openai logo is for openai… I think they want to associate chatgpt with a circle.
@BorisMPower Just as I was reading this tweet I got a “liftoff!” notification from SpaceX
respuesta
@danielsateler1
@iruletheworldmo @de_odat I will never unfollow 🍓
respuesta
@danielsateler1
@elonmusk based or biased?
That’s really the first message of the conversation with no system prompt?
respuesta
@danielsateler1
@sama @DaveShapi Sam replying David Shapiro!! So cool
respuesta
@danielsateler1
@marcelo_earth Jajajajaja
respuesta
@danielsateler1
“Beñch”
@danielsateler1
El proyecto LATAM-GPT está bien para educar y generar cultura de research, pero más allá de eso no va a tener aplicaciones reales. Se va a quedar en la nada.
Si en vez de “otro modelo más”, hicieran un benchmark robusto para evaluar la representación cultural latinoamericana,…
@marcelo_earth Yeah, that’s been around for a while, but it hasn’t really worked.
People have still been complaining about it since I posted in the community forum over a year ago. https://t.co/jnQE7YGV2F
@danielsateler1
Uh-huh, uh-huh, uh-huh, mmm
If I, if I, if I, if I, if I
B-b-b-b-baby g-g-g
So g-g-g, you, you
If I lose you, If I lose you
Uh-huh, uh-huh
@danielsateler1
RT @de_odat: And we live to see another day. The air hums with opportunity. GM. https://t.co/Ums6Of8DYS
respuesta
@danielsateler1
Esta prueba no ha cambiado en años, aun así pareciera que no hay suficientes datos en internet como para que ChatGPT haya podido aprenderla.
Si la prueba requiere pura memoria y no es ampliamente accesible desde internet, tampoco debería sorprender que ChatGPT ni los humanos…
@danielsateler1
Le pedí a ChatGPT Operator que diera la prueba de manejo chilena en https://t.co/5suLqmBTgv. Impresionantemente se metió y fue pregunta por pregunta, cerrando pop ups y respondiendo todo.
Resultado: 29 de 35 puntos, no logró pasar.
¿Cagazo de ChatGPT? Nope, el problema es…
@danielsateler1
consider lighting
@danielsateler1
RT @IterIntellectus: watching sports is *by definition* NPC behavior
@danielsateler1
Have you felt AGI at some point?
@danielsateler1
@ID_AA_Carmack any updates on your agi project?
@danielsateler1
RT @yacineMTB: you go to a public school, staffed with normal people. you learn from normal teachers, trying to teach students to be normal…
@danielsateler1
Michael Reeves is fucking genius
respuesta
@danielsateler1
@GaryMarcus @bakztfuture @yudapearl breaking
@danielsateler1
RT @npew: Existing models failing at your task? Just drop a benchmark and watch the LLM providers trip over themselves to beat it. Problem…
@danielsateler1
Las paredes nunca son planas. Es imposible que una pared sea plana.
respuesta
@danielsateler1
@austinc3301 I think he’s still using text-davinvi-003
respuesta
@danielsateler1
I guess openai is focusing on unifying everything into one model to rule them all—merging the o and gpt series.
But even then, there will still be a lot of models.
@danielsateler1
I’m surprised chatgpt still doesn’t have a model router. With so many models (and terrible naming), we need an AI that automatically picks the best model and tools for the job.
respuesta
@danielsateler1
@kimmonismus xd
respuesta
@danielsateler1
@BorisMPower @apples_jimmy Today I tried operator for the first time… I felt it, once again
@danielsateler1
fucking google. ok.. i will pay you for more storage, i need my fucking mail
respuesta
@danielsateler1
@kimmonismus where do you think o4-mini will be placed in that graph?
@karpathy I get goosebumps almost every time I use composer
respuesta
@danielsateler1
@karpathy +1
@danielsateler1
Why tf do most benchmarks only test raw LLMs?
We should also normalize testing them with tools—browsing, running code, using APIs.
AGI is not meant to work in isolation. Give them the tools, then test them.
@danielsateler1
si empiezo tuitear y no paro para siempre
respuesta
@danielsateler1
@apples_jimmy AVM with phone screen sharing isn’t that great right now… Tried using it to help me pick hands in Balatro, but it keeps hallucinating the cards. Also the model’s intelligence isn’t good enough. If it could use o3-mini as a tool/function, that’d be awesome.
btw I really really…
@danielsateler1
memento aha
@danielsateler1
RT @karencarvajal_o: "el más mejor país de Chile"
No te quejes, chileno, que duermes con la frente en las cumbres, allá donde la nieve bes…
respuesta
@danielsateler1
@cacosat @IterIntellectus we are 40min away from:
“Not so fast at advanced reasoning but still fast”
RT @papmaik1: EL AMOR UN MISTERIO KOMO EL INGREDIENTE SECRETO DEL PAP
@danielsateler1
You will see most people only once
@danielsateler1
RT @de_odat: @DaveShapi I get excited for Dave's videos like I got excited for Christmas eve when I was a child
@danielsateler1
Hubo tierra, agua, sangre, flores Todo eso y también tiempo
@danielsateler1
RT @AndrewYNg: Today's "DeepSeek selloff" in the stock market -- attributed to DeepSeek V3/R1 disrupting the tech ecosystem -- is another s…
@danielsateler1
RT @de_odat: I’m not thriving, but I’m aggressively not giving up.
respuesta
@danielsateler1
@cacoos 4o*
respuesta
@danielsateler1
@cacoos I just tried using the / command again, and this time it worked perfectly!
But somehow, yesterday it didn’t work when I tried it.
Despite the command not working, it’s still weird that o1 didn’t use the canvas even though I made it super clear it should. https://t.co/HwBuppRhHm
@danielsateler1
urf
@danielsateler1
I find it very difficult to make ChatGPT use Canvas for react and html code.
Usually, it doesn’t use Canvas, and if I explicitly say “use Canvas,” it replies “Sure, I’ll use React and <canvas>…”, confusing the Canvas tool with the HTML element.
Am I doing something wrong?
@danielsateler1
I think we’ve, at least, reached a point in time where no one can still believe LLMs are just stochastic parrots…
respuesta
@danielsateler1
@ygrowthco @Yoshua_Bengio @AndrewYNg “we are agents”
respuesta
@danielsateler1
@yacineMTB Been saying that since gpt-4… I think avg people (npc) will agree we have agi later this year.
After this year, we’re not longer “early”.
respuesta
@danielsateler1
@AndrewMayne @NickADobos gpt is surprisingly good at reading and generating base64
They told me it’s a limitation with Whisper and suggested a workaround (that I was already using)—but it’s inconvenient. I also proposed a fix, and while they liked the idea and assured me they’d pass it to the product team, they also said it wasn’t something they could…
I’ve opened a ticket and posted on the community forum, where it seems lots of people are experiencing the same problem. https://t.co/grrBb2GDJk
@danielsateler1
RT @SpaceX: Liftoff of Starship's seventh flight test. The Super Heavy booster utilized flight proven hardware for the first time, reusing…
@danielsateler1
RT @c_valenzuelab: You watch as the old world of media crumbles before your eyes, this behemoth built on scarcity and impossible budgets, o…
respuesta
@danielsateler1
@yacineMTB Indeed
respuesta
@danielsateler1
@emollick 👀2
respuesta
@danielsateler1
@ashleevance “Vance also has a book in the works on OpenAI, for which he said he’s already sold the movie rights.”
@danielsateler1
To write my custom instructions (for 4o), I had to be very specific—almost overly detailed—to get it to behave exactly how I wanted. But with o1 being so good at following instructions, that same level of specificity backfires, making responses feel way too exaggerated.
@de_odat That’s a humble take, feels very acosmist in nature.
Respect.
@danielsateler1
It’s 2025, and doctors still don’t use AI to assist in diagnoses… That’s negligence.
A doctor without AI is like a pilot without radar. In a world where 1.56M people die each year due to diagnostic errors, not using AI is irresponsible.
It’s 2025, and doctors still don’t use AI to assist in diagnoses… That’s negligence.
A doctor without AI is like a pilot without radar. In a world where 1.56M people die each year due to diagnostic errors, not using AI is irresponsible.
@austinc3301 we really need them now more than ever
@danielsateler1
Shampoon't
respuesta
@danielsateler1
@SSSSS7MMMMM @TheAppleDesign @GK3 That means that by default (without goodlock) you can’t do that?
@danielsateler1
Velocity : Design : Comfort.
@danielsateler1
Velocity: Design: Comfort.
respuesta
@danielsateler1
@TheAppleDesign @GK3 One thing I miss from Android is the ability to control individual audio volumes for things like calls, notifications and media. On iOS, you can’t do that, but honestly… I haven’t needed it as much. Apple’s seamless integration kinda makes up for it.
Frustratingly impressive.
respuesta
@danielsateler1
@Rainmaker1973 I reaaaally wanted one back in 2013… heck, I still want one now!
respuesta
@danielsateler1
@512x512 Voice input. What truly matters to me is having excellent transcription—something like Whisper, which is absolutely game-changing.
Advanced Voice Mode is super cool, but what I’m really looking for isn’t voice output. It’s about being able to send long prompts comfortably.…
respuesta
@danielsateler1
@yacineMTB when you know what problem to solve and what questions to ask, you can realize current AI is AGI
