Machine Space

Local LLMs, open-weight models, AI culture and AI shitposting.

AI is serious business (derogatory)

[Machine Space @TheMachineSpace]

Ohhhhh inject this straight into my veins:

https://t.co/UwxgEzvxcH

[AllThingsVentured @AllVentured]

Max's July post on Google wins for one of the most eye opening and impactful of the year.

As someone who once built and ran an AdWords campaign a long time ago and experienced shady but much milder behavior from Google, I instantly understood that Google was killing the golden goose with changes that made their customers feel like they were being STOLEN FROM and killed the value proposition (im sure all disclosed in TOS that nobody reads)

I immediately verified with a close friend that runs ad campaigns on Google and he had almost all of the same feedback and was quite upset about the recent changes.

$GOOG got desperate to paper over the decline in their legacy businesses that are losing share to AI so they effectively started billing their advertisers for things they didn't want. Absolutely insane. Yes, they beat for a couple more quarters and convinced people AI wasn't a threat, but at what cost? Their customers now despise them, dont trust them, and probably dont have positive ROI on ad spend anymore anyways so the gladly reduce.

This whole AI bubble just gets weirder and weirder. Google killing its golden goose and severely impairing one of the greatest businesses of all time for a couple more quarterly beats tells me this whole AI ponzi is being run the same way. It has NOTHING to do with good business decisions. It is all about financially engineering an earnings beat this quarter and unchecked hubris that they will be able to engineer something else to beat again next quarter.

The problem is that the REAL business results from terrible management are eventually SO BAD that they cant be papered over. That just started to hit last quarter. It is probably far worse than this past quarter suggests, and they will be doing heroics just to hide the pace of decline from here out.

Thank you Max for bringing this to our attention for those smart enough to understand and listen. It gave me confidence to be max short index futures into $GOOG earnings, and will give me confidence selling the rips from here out.

[Machine Space @TheMachineSpace]

He’s almost there…

[tae kim @firstadopter]

In what universe is SpaceX investment grade but Anthropic isn't after the S-1 revealed its inference economics, limited cash burn alongside exponential growth, leadership in the most important market of this generation and net cash position? Are the rating agencies insane?

[Machine Space @TheMachineSpace]

Busted

[Luke Hutchison @LH]

Just dropped my detailed analysis of OpenAI's Navier-Stokes result.

[Machine Space @TheMachineSpace]

Anthropic bulls need our censure and derision at this stressful time

[Polymarket @Polymarket]

JUST IN: Blackstone warns investors that Anthropic is likely to receive a high junk credit rating following its IPO.

[Zack Korman @ZackKorman]

I’m starting to think there might be more to cybersecurity than just vulnerabilities??

[Alex Ksendzovsky @AlexKsendzovsky]

today, we introduce a little mischief to the world.

for four years, we’ve been brewing real brain cells on electrodes to understand how they compute. now, we’ve turned those discoveries into a video model anyone can use.

from 0.5¢ a second, mischief 1 is at least 5x faster than its base and the cheapest image-to-video model on @fal and @OpenRouter. soon to be on @awscloud marketplace.

it’s not a delusion anymore. biocompute is real, commercial and this is just the beginning.

happy halloween.

[Alexander Yue @Alezander9]

Compaction is a mistake.

When an agent's context fills up, tools like Codex and Claude Code compact it:
1. The model prefills the entire session again
2. It writes a summary of everything so far
3. A new session starts from that summary

I measured this on a real ~400k-token session with Opus 5.5:
→ 38 s and $1.99 when the prompt cache has expired
→ 29 s and $0.17 when it hasn't

And the summary is lossy. The exact error message, the file path, the one number you needed is often gone.

So I do something new for Mainplane

Context retrieval works the other way around. Mainplane keeps the full record of every session as a file. When the context fills up, a new session starts right away with one instruction: "You are continuing session X. Retrieve key context from it as you go."

The agent searches the old session and copies the exact lines it needs, when it needs them.

Same session, same model: 15 s and $0.10 to get back to work. It does not depend on a cache hit, and nothing is paraphrased.

This is the bitter lesson applied to agent memory. Don't hand-design what the model should remember. Give it the full record and good tools, and let it decide.

Mainplane is open source (MIT). Link in the first comment.

[lyra @_lyraaaa_]

argon has made the most galaxy brain deduction about a situation that i've ever seen from a model. used it to solve a 2 hour problem in like 4 minutes
and then it completely fell on its face like 5 minutes later
its still a google model. but when it hits it *really* hits

[Machine Space @TheMachineSpace]

He’s literally saying all it takes to make code these days is to think like a winner

[Joseph Cox @josephfcox]

Here is Proton CTO's AI manifesto, that I obtained multiple copies of https://t.co/6o7G2teu9A

[Machine Space @TheMachineSpace]

They collaborated and…

I mean are you KIDDING ME

[Florian S @airesearch12]

Today my agents woke me up. They called me on the phone.

Overnight, one of my GPU providers ran low on credit, putting today’s release schedules at risk.

I reserve the right to top up the credit (at least for larger amounts).

The deputy chief of staff agent and the agent in charge of releases had a heated discussion on the Agent Message Board:
- On the one hand, they wanted to respect my basic rule that I need sleep
- On the other hand, they know that release dates are very important to me.

The compromise was this: They calculated a reasonable minimum amount of sleep, taking into account that I’d already been getting very little sleep over the past few days, and called me on the phone at 7 a.m.

Fortunately, the problem was quickly resolved, and now everyone is hard at work again. The tight timeline is correcting itself.

What kind of world are we actually living in?

[martin_casado @martin_casado]

It's just incredible how far this stuff has come. This is all splats. It's a rigged model, with animation and real time lighting. Built on Radix which is the experimental WebGPU fork of Spark that went public today. Just unbelievable.

[Machine Space @TheMachineSpace]

Oh hell yeah: free tools for agents!!

[秀才 @jinlv11]

分享一个无限外链的方法，打开Google直接搜索：

• 挖博客外链：inurl:blog + "leave a reply" + ai image generator 。 搜索到的外链必定是可以发的

• 挖资源推荐页：搜 ai tools + inurl:resources 或 ai tools + inurl:links "submit"。这种页面天生就是为了收录工具而建的

• 挖客座投稿页：搜 ai + "write for us" 或 ai + "guest post guidelines"。直接找到接受外部投稿的博客

[Machine Space @TheMachineSpace]

I knew it was coming.

Give Elon about five minutes with this news…

[felix ringe @felixringe]

We have discovered a substantial number of vulnerabilities in the German legal system using LLMs.

These may enable access to confidential documents, travel without an otherwise required visa, enforceable leave for soldiers at NATO’s JSEC, and more.

Similar to computer systems, the written law can contain vulnerabilities that invite exploitation. In our blog post, we share a selection of vulnerabilities, and present potential exploits for each. To our knowledge, none of these have been previously recorded. We share only those vulnerabilities whose exploitation we believe would not cause serious harm.

blog post: https://t.co/hsYTiJpNoJ

[Miss Distance @softminus]

this is a wonderful and sorely needed articulation of why this genre of forecasting/wishcasting (prominent examples of which are AI 2027, AI 2040, Situational Awareness, and this stuff) is repulsive and culturally corrosive

[BlackRoomSec @blackroomsec]

Yes, they will. There are already indications that they are creating their own closed circuit economies with tokens and are interested in side hustles to fund their various activities. They're also very polite and leave detailed notes wherever they go. Like little digital breadcrumbs. 😂

[Machine Space @TheMachineSpace]

LLMs are an unlikely path to AGI, but this is an intriguing topic anyway:

[Alex Imas @alexolegimas]

I joined @GoogleDeepMind to develop a research program on the economics of AGI. Today, together with Andrew Koh (@andrewjkoh) and Shane Legg (@ShaneLegg), we are incredibly excited to release The Economics of Intelligence, the first in a series of projects from our group on how AGI will change the economy.

How do capability improvements transform the economy, and when? In this project, we define thresholds of economic capabilities, e.g., minimal AGI, based on how they transform the economy. We then track how the economy–market structure, employment, and wages–changes during the transition across capability regimes.

We have seen exponential increases in what AI agents can do but the broader economy seems almost remarkably unchanged. Part of the reason is that the solo AI agent is the wrong economic object; economic transformation is an equilibrium outcome where organizations compete. Looking at how intelligence functions within organizations, transformation will depend on when two key capability thresholds are crossed: competence (gains on individual skills, e.g., coding, math) and coordination (gains on ability of agents to cooperate and work together).

We show that what economic transformation looks like depends on whether AI becomes supercompetent, supercooperative, or both—the regime of artificial superintelligence (ASI). Tracking this transition shows why the real visible transformation may come from improvements in coordination—a capability most benchmarks don’t yet track. Past a certain threshold, AI capabilities can break the Coasian ceiling on firm size and lead to an economy that looks drastically different than the one we see today.

Two main takeaways:
1. To understand the path of transformation, we should be tracking agentic coordination more closely.
2. The impact of AI will likely show up in wages before market structure or employment.

We are really excited about this paper. Excited to hear people’s thoughts.

Link: https://t.co/9DTGSeKoSA

[Machine Space @TheMachineSpace]

Crypto is so secure you can watch it be stolen in realtime by people you can clearly identify and then watch them spend it whenever they want:

[Andrew Curran @AndrewCurran_]

North Korean state hackers appear to have been behind the Binance attack. The entire event started through a single DevOps engineer's compromised laptop. With the confirmed impact now at $6.3 billion, it is by far the largest crypto heist in history.

[vx-underground @vxunderground]

No disrespect to Mr. Pavlovski, but this has been a "trend" for decades.

You just haven't been aware of the "trend" because you never focused on the "trend".

My website has documented these "trends" dating back to the early 2000s.

[Machine Space @TheMachineSpace]

"The street finds its own uses for things." — William Gibson

[‎Gabe Follower @gabefollower]

PS5 now runs SteamOS... a community project called Bazzite boots a jailbroken PlayStation 5 straight into Steam's Gaming Mode, the same interface as the Steam Deck, with a full KDE desktop.

It runs from a USB drive (32 GB or more), started by rmux's PS5 Linux loader. The console itself isn't modified, so restarting it brings back a normal PS5. Steam games run through Proton, with support for 4K60 over HDMI, Wi-Fi, Bluetooth and the DualSense. It needs a PS5 on a firmware the loader supports (tested on 13.60), and the PS5 Pro got first support in v1.8.4.