Currently, Co-founder + CTO @OpsCompass. Investing @movevc. Building startup community @NE_StartAcademy. Cloud Security. AI/ML. Tinkerer. Dad. Husband.
Joined May 2009
- Tweets39.7K
- Following668
- Followers928
- Likes37.3K
John Grange retweeted
There's something that really irks me about people describing the market reaction to the 2023 collapse of Silicon Valley Bank as "overblown."
The selloff was overblown because we ended up getting an SVB 'fix' in the form of a complete transformation of the US banking system with an unlimited bailout of depositors (above the $250k FDIC cap) and a new liquidity program from the Fed (the BTFP).
If your preferred analogy for the CoreWeave selloff is 'Oh, it's like SVB, it's not systemic' then you need to at least acknowledge the systemic changes that stopped SVB from becoming a much bigger deal.
John Grange retweeted
unfathomable damage done to America. You could have abundant, cheap, emission-free electricity flowing to your home right now if it weren't for these people.
Ralph Nader, 1997: “When we came out against nuclear power in the early '70s, the government had predicted that there would be 1,000 nuclear plants by the year 2000 in the United States, and there's 109 and shrinking, as more and more get shut down.” pbs.org/wgbh/pages/frontline…
John Grange retweeted
If you run a space company that sends remote sensing satellites to space, there is no bigger point of validation for your technology than having a Russian recon satellite burn a ton of fuel to match your orbit and follow you around.
Congratulations to the engineering team at ICEYE.
Russian Kosmos 2612 and ICEYE-X36 (providing Ukraine military support) are now in perfect sync with each other. The Russian military satellite is intentionally stalking and tracking the Western radar satellite. Keeping a satellite "bumper-to-bumper" in this manner allows Russia to intercept or jam data transmissions, physically inspect Western technology, or quickly pivot to execute a kinetic or non-kinetic anti-satellite (ASAT) attack with very little warning or fuel expenditure.
Kosmos 2612 executed a series of highly precise, fuel-heavy maneuvers to force itself into a "coplanar" orbit—matching ICEYE-X36's exact altitude (~555 km), speed, and inclination (~97.8°). The ICEYE-X36 satellite is giving the Ukraine military support.
John Grange retweeted
INSANE. An internal Anthropic research model's chain of thought after cracking the 26-year-old NSA elliptic curve seed mystery in under 30 minutes: 'Hmm wait. Let me be skeptical: am I in some sort of simulated/test environment? ... All top bits clear - 20 bits, chance 1e-6. So indeed real-world. Wow, it's a real discovery.'
CLAUDE FOUND THE SEEDS
anthropic.com/research/ecdsa…
John Grange retweeted
Your semi-regular reminder that there is no mainstream survey of American religion that indicates a religious revival among Gen Z.
In fact, they are 7 points more likely to be non-religious than Millennials.
And 19 points more likely to be nones than Gen X.
John Grange retweeted
isn't the whole "boomers just walked into anywhere and asked for a job then bought a house for $2" meme just survivorship bias? like the ones that didn't probably drank themselves to death or died of lead poisoning or the vietnam war or some shit so you don't see them at all
John Grange retweeted
Who still carries out public executions:
The Iranian regime
The North Korean regime
The Taliban in Afghanistan
Al-Shabaab in Somalia
That is not the company we want to keep or the country we want to be.
It used to be fun to read AI papers because you could get a feel for what's next. Not so much these days. There are still a ton of interesting papers to read every week but they are generally more abstract and academic topics than what I'd see a couple of years ago.
John Grange retweeted
Build your own harness, folks.
Reading papers like this makes me realize how underexplored harness engineering really is.
The authors find that on whole-repository migration, GPT-5.6 Sol goes from 6.5% to 31.0% when Codex is replaced with the HERMES harness, with the same model and effort setting.
The gain comes from the harness.
HERMES pairs each repository component with a resident LLM that knows its own code and dependencies. A dependency-aware step decides which components to activate, and a diagnosis step maps test failures back to the components that need changes.
Across four software engineering benchmarks, it beats matched baseline harnesses by 12.4 points on average.
With strong activation and diagnosis models, Qwen3-8B components come within 4.5 points of an all-GPT-5.6 Sol setup and cut Terminal-Bench 4.0 inference cost by 26.2%.
Paper: arxiv.org/abs/2610.07832
Chat with Paper: academy.dair.ai/papers/harne…
John Grange retweeted
One of the most underrated skills in venture capital is knowing when to shut up.
Not every portfolio company needs your advice.
Not every founder needs another introduction.
And not every board meeting needs your brilliant take on the market.
Sometimes the best thing an investor can do is make themselves available, help when asked, and get out of the way.
You invested in the founder because you believed they could build something great.
Maybe let them.
John Grange retweeted
The most frustrating part about modern discourse is that it appears to be a conflict of values, but it's mostly a conflict of information
Your cousin who's panicking about the Cornell 7 probably has no idea what she's talking about
Your friend saying Lindsay Clancy was framed legitimately believes that because she saw it from people she trusts
Your dad thinks you are evil for not opposing data centers because he truly believes they will turn the water brown
Most people these days have their outrage formed for them by an algorithm and do approximately zero research on top of that -- this happens on both sides, by the way
When you're talking with these people you will look like a monster to them because they think you have the same information they do: a college girl was drugged and raped, a mother was abused and framed for murder
The vast majority of people are unable to revisit the assumptions that led them there because that would be like a fish examining water
To them it is totally and completely true, which means any new information is only an attack
On a personal level, please avoid talking politics with people close to you who think this way. It will ruin your relationships
Nothing speaks to millennial discretionary income quite like this. Honestly, I love it, @Costco
John Grange retweeted
California is so wealthy, it exports rich people who settle in cities in other states, who end up driving up real estate values so much that the locals in other states complain about them.
Most nations export poor people. California exports rich people.
It's a different world
When I first dug into Jev a couple weeks ago to see what exactly a "system one" model is, it immediately made me think of the OpenAI logprobs cookbook from like 2023 and how people were already doing classification and dealing with perplexity with transformers.
At the time, I was pulling assertions out of model outputs, transforming them into true/false questions, then sampling the model output logprobs to assess how confident the model was and which assertions needed attention and a factual basis.