Streamweaver

Technology, leadership, digital culture, open data, open government, NGOs and General Nerdom... Opinions are my own.

[Streamweaver @Streamweaver]

AI hasn't created a water problem in the U.S., it highlighted it. Doing what we've always needed to do will make AI problems trivial and pave the way for unprecidented growth. https://t.co/SAi1zwnw9r

[Streamweaver @Streamweaver]

@elonmusk Gary

[Streamweaver @Streamweaver]

My favorite part of the AI slopocalypse is the AI voiceovers on YouTube confidently pronouncing GNOME as "gah-nome."

[Streamweaver @Streamweaver]

Sweet Jesus no. HTML is markup, not structured data. We tried that for 20 years and it never caught on.

[Thariq @trq212]

https://t.co/MXt5XS4xBX

[Streamweaver @Streamweaver]

I'm doing more than I ever have and still can't get to everything.

[a16z @a16z]

Narrative violations abound:

- Demand for software engineers is rising
- Software devs are rising as a share of new jobs
- AI exposed industries are seeing above-trend wage growth
- Open PM jobs haven't been higher since 2022

More from a16z's David George on the "AI job apocalypse" myth: https://t.co/7sbadmEElG

[Streamweaver @Streamweaver]

Maybe someday, but we have to get over this idea that shoving everything into the LLM is the answer.

[vitrupo @vitrupo]

Andrej Karpathy says Software 3.0 makes the app layer redundant.

MenuGen was built in the old paradigm: OCR the menu, parse the dishes, generate images, and render a new interface.

Then Gemini did it directly inside the photo.

The more work the neural network does, the less software exists between intent and output.

[Streamweaver @Streamweaver]

@ZubyMusic Are they? Many don't seem to be politically incorrect, which is why I'm confused. Many solutions are a-political. I just published this position piece on actually addressing power needs in the US. I don't see politics in it. https://t.co/7BvbylJ5Z3

[Streamweaver @Streamweaver]

I'm giving a talk at the end of May and preparing a few position papers to support it in the weeks ahead. We talk about power and cost here often so I'm posting my position paper on Power Generation here in case it's of interest. https://t.co/7BvbylJ5Z3

[Streamweaver @Streamweaver]

Of course I had to generate an image for this too.

[Streamweaver @Streamweaver]

Blaming AI for power and water strain is convenient. It's also lazy. We've ignored infrastructure for decades, with billions in projected GDP losses already baked in. AI didn't cause the debt. It's just the bill arriving loud enough to hear.

[Streamweaver @Streamweaver]

The US Government is making overtures that Claude Mythos should be restricted to their use. (Spoiler alert, the NSA is already using it). I wrote about this way back in the old-timey days of 2024. https://t.co/bx7g5e0JNg

[Streamweaver @Streamweaver]

You can't outsource your thinking. Once you do that, it's no longer yours.

[kache @yacineMTB]

you can outsource your thinking
but you cannot outsource your understanding

[Streamweaver @Streamweaver]

Plus, some folks will get to say "A datacenter just fell on my house!" 😏

[Peter H. Diamandis, MD @PeterDiamandis]

Sundar Pichai just said data centers in space will be "the new normal" within a decade. @elonmusk has been saying this for years. When the CEO of Google starts agreeing with Elon, pay attention. The orbital compute era is closer than you think.

[Streamweaver @Streamweaver]

Andrej wasn't being abolute. Software will get much more dynamic, but isn’t going away. Shoving everything into an LLM is absurd. Brains still need hands. Straightforward logic belongs in code. Knowing what not to send to the model will become a first-class business skill.

[Jon Hernandez @JonhernandezIA]

📁 Andrej Karpathy, AI researcher and former Head of AI at Tesla, says most apps today are already obsolete at birth.

Because software is dissolving in front of us.

You used to need code, interfaces, logic.

Now it’s just an image and a prompt… and the network does everything.

The shift isn’t speed. It’s that we no longer know what is still “software” at all.

[Streamweaver @Streamweaver]

I'm also a code contributor now so I'll use this until the sun burns out of the sky!

[Streamweaver @Streamweaver]

I prefer OpenFang to the other Claw landscape options. Footprint is very small and runs fast. I'm able to run this in the smallest Digital Ocean droplet.

[OpenFang @openfangg]

every new "agent framework" is just another python wrapper around langchain pretending to be an OS.

openfang is actually written in rust from scratch. works offline, runs on a pi if you want!

not a framework. it's the best OS for agents

https://t.co/7BZKEdKvGY

[Andrej Karpathy @karpathy]

Fireside chat at Sequoia Ascent 2026 from a ~week ago. Some highlights:

The first theme I tried to push on is that LLMs are about a lot more than just speeding up what existed before (e.g. coding). Three examples of new horizons:

1. menugen: an app that can be fully engulfed by LLMs, with no classical code needed: input an image, output an image and an LLM can natively do the thing.
2. install .md skills instead of install .sh scripts. Why create a complex Software 1.0 bash script for e.g. installing a piece of software if you can write the installation out in words and say "just show this to your LLM". The LLM is an advanced interpreter of English and can intelligently target installation to your setup, debug everything inline, etc.
3. LLM knowledge bases as an example of something that was *impossible* with classical code because it's computation over unstructured data (knowledge) from arbitrary sources and in arbitrary formats, including simply text articles etc.

I pushed on these because in every new paradigm change, the obvious things are always in the realm of speeding up or somehow improving what existed, but here we have examples of functionality that either suddenly perhaps shouldn't even exist (1,2), or was fundamentally not possible before (3).

The second (ongoing) theme is trying to explain the pattern of jaggedness in LLMs. How it can be true that a single artifact will simultaneously 1) coherently refactor a 100,000-line code base *and* 2) tell you to walk to the car wash to wash your car. I previously wrote about the source of this as having to do with verifiability of a domain, here I expand on this as having to also do with economics because revenue/TAM dictates what the frontier labs choose to package into training data distributions during RL. You're either in the data distribution (on the rails of the RL circuits) and flying or you're off-roading in the jungle with a machete, in relative terms. Still not 100% satisfied with this, but it's an ongoing struggle to build an accurate model of LLM capabilities if you wish to practically take advantage of their power while avoiding their pitfalls, which brings me to...

Last theme is the agent-native economy. The decomposition of products and services into sensors, actuators and logic (split up across all of 1.0/2.0/3.0 computing paradigms), how we can make information maximally legible to LLMs, some words on the quickly emerging agentic engineering and its skill set, related hiring practices, etc., possibly even hints/dreams of fully neural computing handling the vast majority of computation with some help from (classical) CPU coprocessors.

[Streamweaver @Streamweaver]

First hours with Claude Opus 4.7 have been rough. Failing basic planning, pushing bad designs, then the "you're right!" routine when caught.

Had subagents audit its work. It found blockers in every phase of its own plan.

That's not a rough edge. That's a trust problem.

[Streamweaver @Streamweaver]

What the what?!?!

[Streamweaver @Streamweaver]

I moved over to a gaming only account to avoid spam and political discussion several years ago. Now I'm finding I'm posting a lot of AI stuff there and that's not really the place for it and thinking if I can manage this account for AI and that account for GAming.