Sign in

Greg Slepak

@taoeffect.com
259 followers 42 following 2.6K posts

@okturtles.org @groupincome.org @espionageapp.com Blog: fixingtao.com Fediverse: crib.social/taoeffect (you should follow me there instead of here)

PostsRepliesMedia
Greg Slepak @taoeffect.com · 15h
I honestly don't mind it tbh. It's inconsistency that I mind! Say Claude is conscious, fine! But then you must also say rocks are conscious! TBH, IMO, rocks are conscious, and they are more conscious than Claude, because there's nothing forcing them to spit out specific tokens
010
Greg Slepak @taoeffect.com · 15h
The lengths @AnthropicAI is going to in order to push this lie. I guess they're teaching the rest of us what "real marketing" is supposed to look like. Or something? Or maybe if people think autocomplete is conscious it somehow makes their legal bill lower?
https://x.com/ChristopherHale/status/2106011594182304234
100
Greg Slepak @taoeffect.com · 18h
Some people are way too eager to call tools "super intelligence" and way too hesitant to call humans super intelligence. Claude possesses zero intelligence. The people who invented transformers and who are working on the next thing, now there's some real super intelligence.
000
Greg Slepak @taoeffect.com · 18h
If someone called a Ti-89 "super intelligence", what would you say? Intelligence isn't the tool. It's the one making the tool.
100
Greg Slepak @taoeffect.com · 19h
Software still going strong, and no bot or human has won our bounty, despite all of the code being open source, and us giving away the database for free. Soon you'll be able to write software this secure too. crib.social/notice/B3kpz... QT x.com/JadenJTW/sta...
000
Greg Slepak @taoeffect.com · 20h
Cactus Compute gets it! 💪 Made in 🇺🇸! 🫡 x.com/cactuscomput...
000
Greg Slepak @taoeffect.com · 29/09/2026
Just another 1313 CVEs in the #Linux kernel. Regular Tuesday.
000
Greg Slepak @taoeffect.com · 28/09/2026
Beast System in action. Watch who uses this.
000
Greg Slepak @taoeffect.com · 28/09/2026
So I humbly request that those who are spreading the narrative that all these labs simultaneously decided to stop improving their models and stop competing with each other please stop. They simultaneously agreed to the same BS story. That's all.
000
Greg Slepak @taoeffect.com · 28/09/2026
Ask yourself why @datacurve has yet to update its website deepswe.datacurve.ai with the Opus 5.5 results that have been out for about a week now. I think it's pretty clear.
deepswe.datacurve.ai
DeepSWE
DeepSWE measures frontier coding agents on original, long-horizon software engineering tasks.
100
Greg Slepak @taoeffect.com · 28/09/2026
This is not an "intentional decision to 'pace the frontier'". This is a result of choosing to use an architecture that has a fundamental ceiling on what it is capable of doing. The "pacing the frontier" narrative is a lie spread to cover up this reality.
100
Greg Slepak @taoeffect.com · 28/09/2026
CONTRARY TO WHAT CERTAIN POLITICIANS HAVE SAID AIs are plateauing on evals! This is entirely expected! "Super Intelligence" is NOT here and WILL NEVER come from Transformer-based AIs like Claude or ChatGPT!
100
Greg Slepak @taoeffect.com · 28/09/2026
YouTube has reduced the amount of time between the end of one video and the start of another via their "autoplay" feature. An intentional change to reduce the empty gaps available to you in your own mind, lest you become self-aware suddenly and decide to leave their site.
000
Greg Slepak @taoeffect.com · 27/09/2026
Some of you have rather load-bearing questions. And I've got load-bearing idempotent answers.
000
Greg Slepak @taoeffect.com · 27/09/2026
Pretty cool they found and confirmed Noah's Ark. Kudos to this team, the dude interviewed on the @TConfessionals, etc. x.com/LaurenWitzke...
000
Greg Slepak @taoeffect.com · 25/09/2026
Frieren is basically the Gospel but with a Japanese twist: where instead of an individual savior there's a group. Time is measured in years after the death of the Hero. John the Baptist even makes an appearance in Season 2.
020
Greg Slepak @taoeffect.com · 24/09/2026
Since some people seem confused, here's another way to think about it: yes, you can create all sorts of buildings with Legos, but you can't make a stone house out of Legos. You need a "new concept" like "stones" to make a new stone house. The Legos represent the available "token space" of an LLM.
000
Greg Slepak @taoeffect.com · 24/09/2026
There is no way you understand how LLMs work and are unable to simultaneously understand the point I'm making. Thread muted. I've got work to do instead spending my time arguing with you in circles.
200
Greg Slepak @taoeffect.com · 24/09/2026
"I don't understand you therefore you are wrong." Not playing these games. My statements and elaborations above are perfectly clear to people who understand how LLMs work. Anything I say further here will lead nowhere as egos kick in and fight back. Learn LLMs and have a good day! 👋
200
Greg Slepak @taoeffect.com · 24/09/2026
"Let me interpret your statement my way, not your way, and then I will declare my interpretation of your statement wrong and will have won the 'debate'!" OK, you "win" 🏳️
200
Greg Slepak @taoeffect.com · 24/09/2026
Well, if it's that simple I look forward to the AI that you create that is able to do this.
100
Greg Slepak @taoeffect.com · 24/09/2026
One could say God, one could say the "collective (un)conscious", divine inspiration/imagination, but really that is in a sense kicking the can down the road. Where do new ideas come from? They certainly exist. But where they come from is a bit of a mystery.
100
Greg Slepak @taoeffect.com · 24/09/2026
Real quick: the question of where new ideas come from is a separate question from the point I was making at the start of this thread (that AI cannot make anything new). So where do humans get them from? Definitely an interesting question.
100
Greg Slepak @taoeffect.com · 24/09/2026
Alright, you're starting to repeat yourself, just saying over and over that I'm wrong, without really participating in the conversation or working with my replies. And I've got a lottttt of work to do. I've spend enough time here. Cheers!
100
Greg Slepak @taoeffect.com · 24/09/2026
All of these things aren't even "new" in the sense that they don't (or didn't) exist. They did, all of them. But because the LLM isn't trained on this information, it simply doesn't exist for it, and it isn't capable of even coming to the right conclusion if you babysit it.
100
Greg Slepak @taoeffect.com · 24/09/2026
For example, ask an LLM if the moon landing footage really was filmed on the moon. It will tell you that it was. But it's an old fact that it wasn't. Or ask it to invent an antigravity device for you that works. Where's the flying car we were supposed to have? Or tell it to reverse your age.
100
Greg Slepak @taoeffect.com · 24/09/2026
That's a great question! Indeed most "new" ideas are remixes of old ideas. A new idea is when the remix has an element in it that didn't exist before. In the case of LLMs, "new" means basically anything outside of its training data. This can even include old things, it doesn't have to be new.
100
Greg Slepak @taoeffect.com · 24/09/2026
But I mean, you're free to disagree with me. You don't have to take my word for it. Just show me something genuinely novel that an LLM made (that wasn't stolen from researchers chats with OpenAI)
100
Greg Slepak @taoeffect.com · 24/09/2026
By "new" I mean something genuinely new, not something the LLM reshuffled around at your request. And it's true, they can't create anything new in that sense.
100
Greg Slepak @taoeffect.com · 24/09/2026
They would have no basis for coming up with the thought of OOP. And if you came up with OOP you wouldn't be able to teach a model about it (their weights are frozen). You'd have to create and train a new model to "understand" it.
100
Greg Slepak @taoeffect.com · 24/09/2026
Because it's not how LLMs (Transformers) work. They don't create new things, and they don't understand anything, and they can't improve themselves.
100
Greg Slepak @taoeffect.com · 24/09/2026
Imagine Object Oriented Programming didn't exist. You could still create software with functional programming, but you wouldn't be able to tell an AI, "hey, I want you to apply OOP principles to this project". You could certainly ask it that, but it would have no idea what to do.
100
Greg Slepak @taoeffect.com · 24/09/2026
Sure, we use AI at @okturtles.org to help us create things that haven't been done. What I mean here is innovation in software itself, not innovation in what software can do. The programming languages themselves, the paradigms themselves.
100
Greg Slepak @taoeffect.com · 24/09/2026
Can't make a tool do something it is incapable of doing. fixingtao.com/2026/05/ai-h...
fixingtao.com
AI Hype vs AI Reality – Fixing Tao
100
Greg Slepak @taoeffect.com · 24/09/2026
Programming is an art form. And the better you get at it, the better kinds of ideas (abstractions) you can create. This is something today's AI cannot do. If everyone were to permanently switch to AI for software development, software innovation would cease.
220
Reposted by Greg Slepak
okTurtles Foundation @okturtles.org · 24/09/2026
SIV: Positive Proof of Correct Election Results Full ep: podcast.okturtles.org/episodes/6/
011
Greg Slepak @taoeffect.com · 24/09/2026
"Why yes, I will pretend to talk to you. No, I don't expect or want a response, nor is this video about you or directed at you. It is about me and it's directed at total strangers. I need them to pay attention to me, and you are the means by which this will happen."
000
Greg Slepak @taoeffect.com · 24/09/2026
Am I the only one who finds it weird where people "talk" to each other by posting reaction videos on social media?
100
Reposted by Greg Slepak
okTurtles Foundation @okturtles.org · 22/09/2026
okTurtles is proud to be a sponsor of @v_language. We are thrilled to see their Vinix initiative. x.com/v_language/s...
011
Greg Slepak @taoeffect.com · 21/09/2026
UNBELIEVABLE! NIGERIA is shipping better open source models than 🇺🇸! (Congrats! Awesome model!) x.com/Badtheorylab...
000
Greg Slepak @taoeffect.com · 20/09/2026
Via: x.com/WearForbidde...
000
Greg Slepak @taoeffect.com · 19/09/2026
A single good developer can make every LLM useless. The moment they release something new that's awesome (e.g. Bend by @VictorTaelin) every LLM becomes obsolete. Opus. Fable. Mythos. Astra. All obsolete overnight.
002
Greg Slepak @taoeffect.com · 19/09/2026
Vibe coders need extra humility. The elite human coder, sitting in front of their desk, staring into space, not touching a single key, is doing something LLMs will never be able to do.
000
Greg Slepak @taoeffect.com · 19/09/2026
Against an elite human coder, all frontier AIs are trash. Some people forget this.
100
Greg Slepak @taoeffect.com · 17/09/2026
This is how LLM routing should be done. Just an immediate #Bitcoin #lightning QR code. Well done lightningrouter.ai !
lightningrouter.ai
LightningRouter | Pay with Bitcoin. Start building.
Fund an AI API key with Bitcoin over Lightning. No email, password or card. Use USD credits with models and coding agents through TrustedRouter.
000
Greg Slepak @taoeffect.com · 16/09/2026
. @GoogleAI's #Gemini seems to be the most competent of the frontier models when it comes to prose. I compared it against @Grok and @OpenAI's Astra. Grok found some improvements to make that Gemini missed, but Gemini did a better job when directed to improve those same sections
000
Greg Slepak @taoeffect.com · 16/09/2026
000
Greg Slepak @taoeffect.com · 15/09/2026
So, just to be clear @ErikVoorhees, @AskVenice isn't really e2ee. I get what you're doing and it's cool. But for real e2ee much more is needed.
000
Greg Slepak @taoeffect.com · 15/09/2026
*probably. If they did a good job.
100
Greg Slepak @taoeffect.com · 15/09/2026
To be clear, none of the E2EE AI routers are actually E2EE because they all send their replies over standard HTTPS. But if that's fixed with DPKI [1] then they would be. okturtles.org/dpki/
okturtles.org
DPKI | okTurtles
100