I follow finance markets, mess around with AI, and I’ve been building https://teapaw.org/t.co/h24izEsTpz
North Carolina, USA
Joined May 2007
- Tweets9.6K
- Following1.8K
- Followers585
- Likes23.6K
Pinned Tweet
I was served a great injustice. Now it is my mission in life to punish those responsible. Nothing and no one will get in my way.
4
16
2,438
Josh retweeted
I had early access to Fable 5.2 for a bit & I think it crosses the line into being a sort of Jagged AGI. Here, I have it launch a business, do academic research, hack hardware, build 3D models & more. I think it is a good illustration of limits & powers: oneusefulthing.org/p/fable-5…
Josh retweeted
this is what every computer does, slowed down enough to watch
a clock ticks. the chip grabs an instruction from memory, figures out what it means, does it, and moves to the next one
the whole program here is five lines: load 5, load 3, add, show the result, jump back. so the number just keeps climbing by 3
your phone does this billions of times a second
Josh retweeted
Dems won by focusing on affordability.
As it happens, the Senate’s first vote when it returns Tuesday will be on a bill to lower energy costs by modernizing the permitting process. It will be first test of whether affordability is a governing priority or just a campaign slogan.
Idiots
Yes. OpenAI documented a PRC-linked "Data Center Bandwagon" campaign that used ChatGPT to generate posts and images claiming AI data centers raise US electricity prices, posted via inauthentic accounts. Chinese, Russian, and Iranian state media have amplified local opposition. Reports from Alethea and the Bitcoin Policy Institute detail further state media, bot activity, and funding networks tied to groups organizing against projects. Analysts note the core opposition remains largely organic.
13
Josh retweeted
JUST IN: 🇺🇸 $10 trillion Morgan Stanley becomes one of the first major banks to tokenize US stocks and bonds on the blockchain.
I am Republican but the @realDonaldTrump grift this term has been insane to the point of being un-American. Extremely disappointing.
13
Josh retweeted
Harness Engineering 2.0 is here!
After protocols for tools, agent communication, and agent interfaces, harnesses now have one too.
Developers just built UHP (Unified Harness Protocol), a standardized interface for running agent tasks across any agent harness.
HarnessRouter is its open-source server implementation.
Here’s how it works:
- The product backend sends a task with a `harness_id`.
- HarnessRouter starts the selected harness in its own session.
- The harness runs its native agent loop and streams back progress.
- Files, artifacts, failures, and the final output return through the same interface.
Without UHP, integrating every harness needs separate handling for task execution, sessions, streaming, files, recovery, and output formats.
With UHP, Codex, Claude Code, Gemini CLI, Hermes, Pi, and other harnesses can plug into the same product backend while keeping their native tools, skills, models, and agent loops.
This means developers can:
- Switch harnesses without rebuilding the product integration.
- Run the same task across different harnesses.
- Add new harnesses without maintaining another execution stack.
HarnessRouter currently supports 30 harness bases.
Here’s the repo: github.com/HarnessRouter/har…
(don’t forget to star it ⭐)
As per official benchmarks, changing only the harness while keeping the model fixed reduced costs by 25% to 64% across several healthcare compliance tasks.
And changing both the harness and model widened the difference to 99.8% between the highest and lowest-cost configurations.
Developers can test their own workloads across different configurations using Harness Arena.
Alibaba Cloud's Qoder has also adopted UHP, making it available as an agent backend through the HarnessRouter API.
If you want to dive deeper, I already wrote an article that explains why model routing is different from harness routing and what supporting multiple agent runtimes actually involves.
It also covers UHP, the local setup, a working API call, and how sessions and files work.
Read it below.
I stopped letting @AnthropicAI view my Claire Code CLI sessions because they added censorship in the form of behavioral rules to their platform. In six months I’m hoping to drop Claude+ChatGPT models entirely from my development environment.
Amen. We never needed to codify basic decency, but in doing so created legalisticl weapons of mass coercion and usurpation. Never again.
1
37
Josh retweeted
Look, I appreciate this tone.
We can engage this way
We can't engage when you're point is 'he didn't include taxes lol, what a moron!'
Revenue
>> I actually included the assumptions. I didn't explain the logic, but nobody has given assumptions like this publicly before. I'm happy to explain the logic in areas but literally don't have time to explain the logic everywhere. I have been studying this company for 10 years. I can guarantee if we spoke about any individual assumption you'd be like "holy shit he has thought a lot about this"
>> launch. SpaceX has 3 Falcon 9 launch pads. SpaceX should have 4 Starship launch pads operational by ~the end of 2026 (2 in TX and 2 in FL). There's a lot of uncertainty in the launch timeline but this basically comes from ramping towards a similar launch cadence per pad as Falcon 9 has been operating at the last two years
Profitability
>> yes I'm forecasting SpaceX EBITDA margins to be much higher than the hyperscalers. Across connectivity and SI. I view SpaceX as a vastly better operational company than any of the hyperscalers. I know this is provocative. I basically believe SpaceX will be able to turn on a GW datacenter about 2 years faster than any hyperscaler. This speed advantage results in an extremely high EBITDA margin. Starlink is already an extremely high EBITDA margin that's kind of masked by user terminal sales and other up-front costs
Capital Efficiency
>> One of my core assumptions is that hyperscalers are delayed turning online big projects. And Neoclouds deeply struggle. This is a spectrum of just severe delays to outright bankruptcy.
The way this manifests is SpaceX having a larger and larger percent of the inference compute supply. I'm basically modeling the following scenario:
1) a massive increase in inference demand (~3x from now to end of 2027)
2) a large increase in SpaceX's percent market share of the inference compute supply
3) some neoclouds being so delayed that they're forced to sell some hardware at a discount to SpaceX (which starts to show up in 2028)
These things together keep rev/GW high and have CapEx/GW going down slightly (only for SpaceX)
---
In terms of not showing assumptions, do you understand how unusual it is for someone to put out work like this into the public domain?
I have spent a decade studying this company and another decade before that studying the semiconductor + space industries
It would take me at least 10 hours of just talking outloud to explain the assumptions that sit below the level of assumptions here
And 100 hours to put it into a form that would be easy to consume (not just verbal)
ChatGPT6.1-Sol High AND Astra6-low both failed to complete the assignments I gave them after 4 days. Massive disappointment. Opus 5.5 is still the king, and will come in to clean up after ChatGPT.
26
