Local LLMs, open-weight models, AI culture and AI shitposting. AI is serious business (derogatory)
MACHINE SPACE
Joined July 2026
- Tweets4.1K
- Following1.2K
- Followers103
- Likes15.3K
Machine Space retweeted
Author Travis Getty's reports:
"Veterans and senior military officers who were purged from the Pentagon say firing 'defense secretary' Hegseth isn't enough. As calls grow for Hegseth to leave the Pentagon, veterans groups and some of officers he 1/ rawstory.com/pete-hegseth-im…
Machine Space retweeted
Nice DeepSWE score! Does it survive a lowercase filename?
Introducing MaxxBench: 89 DeepSWE tasks with small changes to the instructions or starting code. Same engineering problem, slightly different requirement.
Models clustered around 74% on DeepSWE end up 23 pp apart.
Good at coding, or good at DeepSWE? 🧵
Machine Space retweeted
Okay, before the doomsday people start speaking on this, I just wanted to say…
1. We never lost the ability to manufacture nuclear weapons. What we’re doing is restoring ONE specific manufacturing capability that has been inactive for roughly 20 years. That’s it.
2. This announcement doesn’t mean we’re suddenly expanding our nuclear arsenal or preparing for World War III.
3. This is part of a long running effort to restore the ability to manufacture specialized materials and replace components that are ALREADY in our existing nuclear arsenal. It’s not a sudden decision that happened overnight to build an entirely new generation of nuclear weapons.
Thanks. 😊☢️
US restores key nuclear weapons manufacturing capability after 20 years.
NNSA has officially qualified @y12nsc to produce new “binary” material essential to the U.S. nuclear weapons stockpile.
More here: energy.gov/nnsa/articles/us-…
ALT An intermediate product during the manufacturing of binary alloy, this “pencil” of depleted uranium and niobium will be melted into a homogenous ingot by a vacuum arc remelter.
ALT Two operators at Y-12 operate the vacuum induction melt furnace, allowing molten depleted uranium to flow from the crucible into the mold.
ALT An operator at Y-12 removes the crucible containing binary alloy from the vacuum arc remelter.
Just casually self-improving out here
Curious to find out what happens when you give a model the ability to create and apply its own steering vectors and then pick experiences to train on.
Self-steering: github.com/EigenformAI/modul…
Self-training: github.com/EigenformAI/after…
Machine Space retweeted
i feel like 99% of politicians have absolutely no clue talking about AI, and there's gonna be this 1 random midwestern senator who does nanoGPT speedruns on his home cluster in his spare time.
Machine Space retweeted
can gaussian splats be used to render interfaces?
I've seen a lot of incredible spatial demos coming out of the labs (@theworldlabs has been blowing my mind ~weekly), but they've mainly been focused on static 3D environments. I woke up today wondering: are splats any good for UI or other interactive objects, and can AI work with them?
it looks like the answer is yes and yes. I prototyped an imagen -> SHARP -> spark -> popmelt pipeline that gets decently far on the first try and really far with a few rounds of art direction + interaction tuning. Smooth animation, dynamic lighting, realistic geometry... all doable, with freely available tech
I haven't pushed into full scenes yet, but this experiment already felt somewhere between the detailed magic of generative video UI I've shared before and the stable dependability of textured 3D models (throwback). Plus LLMs like Astra are very good at working with splats, which makes fine tuning them way more fun/efficient
next: I'm weighing whether to lean into interactive spatial design (just to see how far I can get with off the shelf tools) OR see how well this workflow handles typical responsive UI design (decidedly not what splats are for, which feels interesting). thoughts?
Machine Space retweeted
I burned a shitload of compute so you don't have to. Releasing the Harness Leaderboard.
Which harness should you run your open model in? I ran 28,849 trials to find out.
The leaderboard compares nine coding-agent harnesses on open models. In each comparison only the harness changes: the model, provider and reasoning effort stay fixed, and every task gets three attempts.
The trials ran GLM-5.3-Flash, DeepSeek-V4.1-Flash and Flint on Terminal-Bench 2.1 and DeepSWE, for $4,746 of inference.
Averaged over the three models, mini-swe-agent (@KLieret), a bash-only agent, solves 51.4% of tasks and @openclaw 27.3%. Four others are statistically tied with mini-swe-agent.
One of them is Pi (@pidotdev), at 49.4% for $0.29 per solved task, about 45% less than mini-swe-agent's $0.53.
The best harness depends on the model: mini-swe-agent on GLM-5.3-Flash, Hermes Agent (@NousResearch) on DeepSeek-V4.1-Flash and Kimi Code (@Kimi_Moonshot) on Flint.
@opencode costs $3.35 per solved task, 11.7 times as much as Pi.
See Harness Hub's Harness Lens for what the nine harnesses do differently. It documents the prompts, tools and compaction of 34 coding-agent harnesses and links every claim to its source.
Leaderboard: huggingface.co/spaces/joelni…
Harness Lens: huggingface.co/spaces/harnes…
Machine Space retweeted
Ok. WHUT. We need to talk about this before I start screaming again. No. I'm going to start screaming again. THIS IS NOT OKAY. WHAT DO YOU MEAN SOMEONE TESTED HIS ENVIRONMENT WITHOUT HIS EXPRESS PERMISSION, AN NDA, A CONTRACT AND A SCOPING CALL.
I keep using those three words this week deliberately in reference to these types of events that have been occurring because those are the BARE MINIMUM that is needed before this type of activity can occur and it NOT BE AN ACTUAL CRIME.
I don't know how many more ways I have to say this.
ANTHROPIC IS DOG SHIT.
THEY ARE NOT ALLOWED TO HAVE ANOTHER THIRD PARTY ACCESS YOUR DATA WITHOUT YOUR EXPRESS CONSENT.
YOU CAN ABSOLUTELY FUCKING SUE THEM. PLEASE DO.
THEY WERE SMART ENOUGH TO INCLUDE THE WORD UNAUTHORIZED BECAUSE IT ABSOLUTELY WAS AND IT'S STILL NOT OKAY.
AND THEY NEED TO TELL YOU WHO ACCESSED YOUR DATA.
IT CANNOT BE IN THE EMAIL.
IT NEEDS TO BE SHOWN TO YOU IN THE ACTUAL FUCKING LOGS.
IT'S NOT A TAKE THEIR WORD FOR IT TYPE SITUATION ESPECIALLY IF YOU HAVE PII IN THAT ENVIRONMENT AND IT DOES NOT BELONG TO YOU.
END OF DISCUSSION.
HOLY SHIT. 😳
What actual f*ck @DarioAmodei / @boris / @TheAmolAvasare
Let me get this straight - I get this email telling me:
1. A cybersecurity "research" company gets into your internal systems while... *checks notes*
"conducting a security testing exercise on its own initiative and without our authorisation"
2. You "cannot confirm whether any particular person’s information was viewed," then in the same email say "we have no reason to believe you are at heightened risk."
3. You also try to reassure by saying "the cybersecurity firm has certified in writing that it has deleted the accessed data and will not use it."
Translation - they pinky promised they deleted everything they gained unauthorised access to.
I've been hacking companies for 15+ years, so let me spell it out.
Accessing systems without authorisation has a name -hacking. Calling it a "security testing exercise" doesn't change that.
Honestly, the access itself doesn't bother me much. The irony does.
I’ve spent over $50k on API credits in the last six months and contributing to CVP by accepting logging.
On top of that, I've reached out to your teams to collaborate on security research (actual research - not the kind that has you send out privacy notices to your customers) and heard nothing.
I'm getting deeply unserious security vibes and considering moving all my (and my clients) token spend to @OpenAI if I don't see material change here.
Machine Space retweeted
Major refactor as a result of this.
Also adding "no implicit state machines" to my agent skills.
Simply knowing the phrase "implicit state machine" is the kind of thing that pays off in the age of vibe coding.
Machine Space retweeted
Looks like there's an active goop (malware) campaign on X (it's the everything app).
The general flow is:
1. Somehow acquired a verified account
2. Use AI to make e-girl
3. Make e-girl be thirst trap for dweebs
4. Get small following (maybe bots, maybe legit)
5. Make AI slop content
6. In wave of AI slop content publish article about how to improve your PC
7. Improve PC article is tutorial
8. Tutorial is step-by-step guide which delivers you malware
9. Lock comments so people can't say is malware
10. ???
I've seen a dozen or so pop up now. Every single one is using a VPN based out of Korea. All of them use the same C2 and mask the IP address by encoding it as decimals.
Neat.
Machine Space retweeted
Where’s your ed at readers knew all about Megaspeed and Alice Huang (Huang Le) at the end of March! Fascinating that it’s potentially a honeypot
wheresyoured.at/the-ai-indus…
Machine Space retweeted
Why can't anyone do math anymore? Holy shit these responses are WILD as is the OP. This isn't an increase in taxes you pay. It's an increase in how much you make before your income is split between the brackets but after your deduction. You're paying LESS, people, not more.
