Replying to
@GrapheneOS@grapheneos.social Still waiting to hear if you support Israel … Have you ever donated to an Israeli cause?
Remote
Greg Slepak
@taoeffect@crib.social
209 Followers
202 Following
50 Posts
Joined May 09, 2024
Homepage:
Keybase:
greg
Replying to
Where does inspiration come from?
Not from an LLM.
New ideas and new concepts, even basic ones, are incomprehensible to the autocomplete machine, and they cannot generate them or use them without baby-sitting.
Replying to
It honestly infuriates me that some people in computer-science adjacent fields still, to this day, entertain the idea that these things are conscious.
They are specifically engineered *not* to be conscious.
Congrats to https://x.com/youssofal_
I ran into issues last time I tried it, but now the latest version is indeed the fastest/best local coding experience on the Mac!
https://mtplx.com
Updated Dez to add:
- Settings panel (currently only Venice supported)
- Full-text search to history
Replying to
Let me give a concrete example:
The AI might try to install a package from the repo that we know isn't there, and if it fails, then try to install it from where we know it's supposed to install from.
This is nonsensical. Just install it from the known location.
Fable BTW.
Replying to
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.
America's domestic priority #1 should be fixing its food system.
Removing all toxic food from schools, grocery and convenience stores.
Then perhaps:
- removing negative foreign entanglements
- open source AI
- improving the transportation system
Replying to
Ask yourself why @datacurve has yet to update its website https://deepswe.datacurve.ai with the Opus 5.5 results that have been out for about a week now.
I think it's pretty clear.
Opensource AI must win.
Build an AI company that has this motto and you will get our $ and win.
UNBELIEVABLE!
NIGERIA is shipping better open source models than 🇺🇸!
(Congrats! Awesome model!)
https://x.com/Badtheorylabs/status/2102046067990692093
Replying to
@duxsco@fedifreu.de @mullvadnet@mastodon.online If the VPN provider does a good job with preventing logging, they are a service that anonymizes your IP.
I grew up in a world where 98% of the political discourse was disconnected from reality.
Today things have shifted quite a lot.
Now it's more like 60% of the political discourse is disconnected from reality.
A serious improvement. To be celebrated!
Replying to
Spammers and SEO hackers are now using LLMs to post perfectly reasonable sounding comments on Wordpress blogs.
Here's how it works:
If your Wordpress blog has a URL field in the comment form, they add their URL there.
You approve their comment and their site rises in the search rankings as a result of the linkback.
It seems like a perfectly legitimate comment but it's AI-generated and the entire point is to SEO hack.
Here's how it works:
If your Wordpress blog has a URL field in the comment form, they add their URL there.
You approve their comment and their site rises in the search rankings as a result of the linkback.
It seems like a perfectly legitimate comment but it's AI-generated and the entire point is to SEO hack.
Replying to
Nobody wants the road proprietary AI leads to.
Open source AI is the only way to go.
We need Open Source models to start *LEADING* proprietary models instead of lagging.
Replying to
I get it, you are lazy.
You want to produce content every day and do not have the inspiration to write it yourself.
At least try this: give the AI several samples of your writing before asking it to write your script, and ask it to write the script in your voice. 🙏
Dez 0.4.0 adds support for simultaneous streaming and rewrites the core to move most of the Rust code into TypeScript.
https://github.com/taoeffect/dez/releases/tag/v0.4.0
https://github.com/taoeffect/dez/releases/tag/v0.4.0
Beast System in action.
Watch who uses this.
Hearing a grown adult repeatedly yell, "Siri, turn up the volume!"
🙉
Replying to
Also, this comment only applies to people who know how to code.
Those of you who don't shouldn't be writing code yourselves, and you shouldn't be using an AI to write code for you, because the AI can code better than you, and that's a blind-leading the blind situation.
Everyone except George Lucas it seems agrees that human hands cannot move fast enough to intercept photons.
Driving me crazy. This song contains a melody that comes from another song that was released perhaps over a decade ago.
What is that song?! Anyone know?
It starts here at around 1:12
https://youtu.be/UNU4nTgGq5k?t=71
djimboh - China Plate
Replying to
All of the LLM CEOs are lying to you.
They're selling you a BS story that LLMs will get better and outperform the best humans.
The only way that can happen is if the best humans die off.
Replying to
What LLMs can do better than the best humans: they can write large amounts of pre-existing code quickly.
Concepts that the LLMs have not been trained on cannot be produced by LLMs.
LLMs are useful only in the hands of a professional coder.
Replying to
@GrapheneOS@grapheneos.social Just a hypocrisy check since you're filling my feed with this drama.
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.
https://okturtles.org/dpki/
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.
Replying to
This isn't ideal. But at least my feed won't be flooded with deep fake avatars who all sound the same.
Hearing someone repeatedly yell, "Siri, turn up the volume!"
Don't do this.
I will accept a locally hosted AI that you don't have to yell at.
But broadcasting to the neighborhood that you're unable to turn a knob and Silicon Valley's data centers to do it for you is just sad.
Replying to
Really it should be based on decible level. They'd be outlawed as a byproduct.
AI vocab:
"Vibe coded": 0% human effort beyond "do the thing". Increasingly sophisticated slop.
"Vibe engineered": Semi-human driven development.
"AI-assisted": Human pays attention to what the AI is actually writing. Intervenes much more frequently & at code level.
"Vibe coded": 0% human effort beyond "do the thing". Increasingly sophisticated slop.
"Vibe engineered": Semi-human driven development.
"AI-assisted": Human pays attention to what the AI is actually writing. Intervenes much more frequently & at code level.
These are the best results I've seen yet for an A3B model. Congrats to the @KwaiAICoder team!
KAT-Coder-V2.5-Dev is now open-weight. Built on Qwen3.6-35B-A3B and post-trained with the proven KAT-V2.5 recipe—from data construction to the SFT/RL pipeline and optimization strategy. 35B total, 3B active. Apache 2.0
Via: https://x.com/KwaiAICoder/status/2080277845700698166
https://huggingface.co/Kwaipilot/KAT-Coder-V2.5-Dev
Replying to
@devnull@mamot.fr @duxsco@fedifreu.de @mullvadnet@mastodon.online It depends on what's meant by "anonymizes".
If the person means "makes more anonymous" as I said it's perfectly true.
For example, Tor also doesn't give you perfect anonymity. It's just more anonymous than VPNs.
Anonymization is a spectrum, it is the process of removing identifiers. Neither Tor nor VPNs do it perfectly. VPNs are more anonymous than a direct connection and Tor is more anonymous than VPN.
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.
My new_session tool in my fork of Crush is superior to Claude Code's /compact thing.
Driving me crazy. This song contains a melody that comes from another song that was released perhaps over a decade ago.
What is that melody?! Anyone know?
It starts here at around 1:12:
https://youtu.be/UNU4nTgGq5k?t=71
djimboh - China Plate
You cannot leave LLMs unsupervised.
You cannot "vibe engineer" unless you are a developer yourself.
LLMs are not magic knowledge machines. They are semi-retarded autocompletion dohickies.
If you do "vibe coding" or anything of the like, you must use version control liberally. Commit every change. The "top of the line" LLMs are retarded and they may do idiotic things like overwrite files and lose work for no reason.
Vibe engineering and vibe coding is a mixture of prayer to a retarded autocomplete engine and gambling.
You must treat it as such.
Sandbox the beast.
Version control every change it makes.
Backup your data.
You cannot "vibe engineer" unless you are a developer yourself.
LLMs are not magic knowledge machines. They are semi-retarded autocompletion dohickies.
If you do "vibe coding" or anything of the like, you must use version control liberally. Commit every change. The "top of the line" LLMs are retarded and they may do idiotic things like overwrite files and lose work for no reason.
Vibe engineering and vibe coding is a mixture of prayer to a retarded autocomplete engine and gambling.
You must treat it as such.
Sandbox the beast.
Version control every change it makes.
Backup your data.