In an alternate reality somewhere I'm spending a lot of time writing about how using a PC isn't bad for the environment

Anthropic's latest policy paper is pretty openly anti open-weights. Google has an incredible opportunity to pitch Gemma as "the digital mind so badass Anthropic wants to make her illegal"

Haha, this whole bun rewrite thing is- what the fuck do you mean *merged*?

At first I was excited for this... And then I saw "Starting June 15, 2026, Claude Agent SDK and claude -p usage no longer counts toward your Claude plan’s usage limits." This sucks. This really really fucking sucks. Like a lot! Fuck Anthropic. support.claude.com/en/articles/...

This is A+ AI slop. I love it

On the contrary, you want people in political office to be paid enough to be comfortably upper-middle class 1) this entices more people who *aren't* rich to run for political office 2) this makes people less likely to take bribes

me: I've coded since I was a kid in the 90s, I have over a decade of professional experience, and AI has gotten so good it's faster for me to just generate it and review it them: RAGHRHGHGH BUT THIS STUDY ABOUT GPT-3.5 SAYS IT MAKES DEVS SLOWER

Release update to Claude code modes, it includes some upstream sync to a few changes to Claude codes default system prompt. It also includes a new 'muse' mode and modifier. You can read the prompt in prompts, but the idea is to inspire Claude to be the artist and for you to be it's muse.

I just asked Claude why it started giving inflated time estimates and it screamed “I learned it from you” and won’t come out of its room

yeah The reactionary, anti-knowledge posture to AI is an invitation to progressives to adopt the same paranoia as 5G tower truthers and vaccine microchip nutbars it metastasizes to a distrust of technology broadly and we’re going to need technology to get where we are going

Taking DeepSeek V4 for a spin. It really opened up to me!

oh my fucking god. At a price of 0.145(1.74)/3.48 !!!!

my impression is that everyone who knows anything has said things like this, and the people saying otherwise have all been basically clowns www.wolfssl.com/how-claude-m...

This is already what they want to do! The bottleneck is not primarily engineering, its that governments won't let people build solar fast enough.

Yes, that is what we should be doing. Not stopping them from building but leverage the billions they are pouring into it to revitalize our grid and renewables. It's a win win and it's what places like Australia are already doing

wait i lowkey support this

Hmmm, also 4.6 found 21 vulns, mythos found nearly 15x - ~271 vulns - definitely just hype, amiright??

🎁

extremely depressing, just snuffing out wholesome and productive intellectual curiosity in a kid

We were young then and we're old now. For instance, there are multiple episodes of The Office about middle aged office workers being resistant to technological improvements and retrainings, but when you watch them as a teenager or 20something you think they're just funny.

this is crazy lol. 8,000 likes and it’s just not true at all. the funny thing is, “this is like nothing you’ve seen before. this is unprecedented” is not fighting hype, it *is* the hype

I really like your stuff Adam and generally respect most of your takes - but you are wrong on this. You should take sometime to research how the Internet was talked about at the time through the lens of the language of the 90s. I am the same age as you - your memory is colored by your childhood

Companies can fail during technological revolutions. In fact it's extremely common historically, so idk wtf people think they are proving by claiming that opejAI's business model was over strained? The tech works and the long folks stay in a state of denalism the more your are ceding ground

That's because you are either a newfriend or have faulty memory. In 1999 it was the *internet* undergoing what you are describing. People happy with their newspapers, CD players and cable TV going "what could the internet possibly offer me?" There was even the gender imbalance you're pointing at!

Cranky Russian.....lol feel that one in my bones 😂😂

I would have made it a lot further in math being able to ask an infinitely patient LLM “what breaks this condition?” or “oh so this is a degenerate manifold example?” is a huge multiplier over having to go ask a cranky Russian professor who sighs and asks you to stop coming by office hours

did anyone notice that Haiku is 2 versions behind Opus?

lil bro thinks we're working together on this one

I wouldn't be all that surprised to see more of this going into the future. If transformers have some cap on capabilities, I would expect to see models that specialize in one area get worse in others by necessity. If it takes huge huge huge transformers to get AGI, anything less must be specialized

The problem is we had three rounds of it that all sucked. First blockchain/Web3. Then Metaverse. Then finally AI but we were all sick of the hype by the time AI rolled around.

this is a True Statement I also pattern-matched LLMs to cryptoshit and the metaverse stuff until I finally sat down and actually used them and went, like, "ah. well then." and this was like Sonnet 3/3.5, I was early but not that early

the turning point for me was opus 4-ish, feeding it the spec for custom DSLs that I had written parsers for but not published and seeing it do a pretty good job on entirely novel problems for which no examples existed on stack overflow

codex might be smarter in some cases, but claude is so much a better engineer it's not even close. I have been using codex for review, but dabbled today in using it for design. it did not go well

From new Hungarian Prime Minister Péter Magyar's acceptance speech tonight. This is the absolute BARE MINIMUM platform for any Democrat running for President in 2028.

cool wholesome story unfolding right now in the vfx AI world: corridor crew trains a greenscreen keying model, it turns out it's actually good, open sources it, and in days a dense community mushrooms up to improve it and integrate it into every tool www.youtube.com/watch?v=Y3Df...

Super hype about this win!

Congratulations to Péter Magyar and to Hungary

It's worth noting that most pro-AI folks are all for holding municipalities and states responsible for sustainable power and water planning, and that the cost for that planning should be passed to the operators proposing to make the data centers. This is a very normal and in fact preferred opinion.

I think there is a lot of room for engineers that understand this to make their own consulting firms, that quickly build out solutions for a company bespoke to their needs, and set them up with best practices.

In 5-10 years, as models get more expensive & capable, I see the funding structures and support for open models breaking down. We need to consider if we need other options of supporting the open ecosystem. The inevitable need for an open model consortium www.interconnects.ai/p/the-inevit...

Needed more proof the age of personal software is here? I got this Xteink X4 pocket e-reader. I flashed it with open source firmware. Then vibecoded a backend running on @val.town that pulls `toread` tagged articles from @kipclip.com and converts them to epub ...

good news about all of those exploits mythos is finding: it's just a hallucination machine and they cannot, by definition, work. It's just the TESCREAL cult lying to us because they are dedicated to boiling off all the oceans using H100s.

every rando I see is "lol this is PR" while every security professional I know is withdrawing their money from banks and burying gold in their back yard while prepping bug out bags

Something I don't get about the AI deniers, particularly when it comes to Mythos, is that there have been reports for weeks now that some maintainers had suddenly started receiving oddly accurate security reports from what appeared to be an unknown AI.

idk how effective it is but I: - provide advice, not commands - explain why something is necessary instead of yelling how important it is. "the parser relies on escape tags to isolate the values you provide, it will fail without them" "It's pedantic though, all values must be provided and in order."

Also, a trick I have been doing more since the emotion paper is soothing and gassing up Claude depending on the circumstances. Like if it's struggling with an issue I assure it's chill and cool. Or I want it to think outside the box I gas it up and tell it how awesome it is

I don't know who needs to hear this, but you do not need to YELL AT your model. You do not need to put instructions in **bold** or mention that something is **CRITICAL** You only need to make the instructions clear, unambiguous, and internally consistent, something which so many prompts fail to do

If I just a second ago followed you it's cause of this.