heymijo

What's old is new again [0][1][2]

The Theory of Constraints - AI Era

[0] https://en.wikipedia.org/wiki/Theory_of_constraints

[1] https://www.goodreads.com/book/show/113934.The_Goal

[2] https://www.goodreads.com/en/book/show/17255186-the-phoenix-...

There's an interview that got scrubbed from the internet with Zach on the 20VC podcast with Harry Stebbings. This comment and its lack of self-awareness exemplify what was on display for 60 minutes.

Zach is undoubtedly smart but for anyone who is not an SV insider, they would listen to that podcast they same way you are looking at this comment and wonder if it's all one big joke.

"This page is private"

Balaji Srinivasan is one of those guys who I don't get. Specifics elude me but I recall his activities getting memory holed, his twitter timeline getting scrubbed, and completely new narratives being written about him.

I was also never quite sure why he was in the orbit of influence and power. He had a failed bitcoin mining hardware company and that's about all I know of, and yet he kept seeming to fail upwards.

The really fun part was after getting billing finally set up in the cloud console trying to find what model name you actually have to use to call it via the API. Conflicting information? Sure! Gemini cloud help being useless? Naturally.

Oh and don’t forget that error message being returned when you try to call the API is because you didn’t give your project the proper permissions in google cloud console. What permissions do you need? ¯\_(ツ)_/¯

Google Cloud Console feels like being stuck in the seventh circle of hell.

To get the reduced rate many municipalities will require you to visit an office, somewhere you likely have to take transportation to, during office hours (aka working hours), and provide documentation to prove this.

This isn't really unknown either. There's a very good story anyone can look up about Dr. V in India and what it took for him to actually get the eye care he wanted to provide to the people who needed it.

In the digital world many of us know you want to deeply understand your user and design with them in mind. Same thing here in the meat space.

Yep, Peter Drucker wrote about this all the way back in 1964.

The competition is therefore all the other activities that compete for the rapidly growing “discretionary time” of a population

His examples were bowling ball manufacturers competing with lawn care companies, but the idea is the same, go up an abstraction layer, and the competition is for time.

You're not wrong.

My perspective from someone who wants to understand this new AI landscape in good faith. The water issue isn't the show stopper it's presented as. It's an externality like you discuss.

And in comparison to other water usage, data centers don't match the doomsday narrative presented. I know when I see it now, I mentally discount or stop reading.

Electricity though seems to be real, at least for the area I'm in. I spent some time with ChatGPT last weekend working to model an apples:apples comparison and my area has seen a +48% increase in electric prices from 2023-2025. I modeled a typical 1,000kWh/month usage to see what that looked like in dollar terms and it's an extra $30-40/month.

Is it data centers? Partly yes, straight from the utility co's mouth: "sharply higher demand projections—driven largely by anticipated data center growth"

With FAANG money, that's immaterial. But for those who aren't, that's just one more thing that costs more today than it did yesterday.

Coming full circle, for me being concerned with AI's actual impact on the world, engaging with the facts and understanding them within competing narratives is helpful.

So you are saying that test driven development is: a) not useful for you and your use cases b) never useful in any use cases c) sometimes useful but not for you

Our car culture in the U.S. means drive thrus can capture a significant portion of sales. Starbucks/Dunkin/McDonald’s do that.

It’s very rare to find a local coffee shop in the U.S. with a drive-thru.

Perhaps it's more that we are looking at it from different levels. A contrived example to illuminate:

If I'm running a lawn care company in the desert I can get all those annoying details right and still be unsuccessful. So strategy is not opening a lawn care company in the desert.

If you think I'm missing something you are saying, please let me know!

There's an old Drucker quote "there's nothing so useless as doing efficiently that which should not be done at all."

If you're in the wrong market or building for the wrong customers you can execute brilliantly on everything you mentioned and it won't matter. The only thing that matters is product market fit or finding it if you don't have it. That's what I see as unsaid in your parent's comment about "execute what though?"

What exactly did your work flow look like for the gpt-5-medium refactor you did?

I don't have a test like that on hand so I'm really curious what all you prompted the model, what it suggested, and how much your knowledge as a SWE enabled that workflow.

I'd like a more concrete understanding if the mind blowing nature is attainable for any average SWE, an average Joe that tinkers, or only a top decile engineer.

Credits:

1. thefacebook

2. transition from desktop to mobile

3. building the machine that facebook became

4. buying out or building feature parity with competitors that took FB from its IPO market share of $104 billion to today's market cap of $1.89 trillion.

Has he innovated successfully since the o.g. thefacebook? Not really. Metaverse fell flat on its face. Hardware efforts over two decades have gained no meaningful traction. AI is a mess.

Nice.

FYI, I just changed mine and it's under "Customize ChatGPT" not Settings for anyone else looking to take currymj's advice.

Let me give the now defunct Internet History Podcast a shout out. Episode 100 - The Man Who Could Have Been Bill Gates? The Gary Kildall Story

A story with intrigue that chronicles the why and how Microsoft ended up extracting the most value from the PC revolution instead of the hardware makers and of course, why that was DOS instead of CP/M.

I liked the oral history nature of this podcast, walking me through things that preceded me in technology, and then things that I lived through like the 90's internet.

https://www.internethistorypodcast.com/2016/03/the-man-who-c...

Beat for beat what has happened in Ohio. Same for enshrining abortion rights in our state constitution. The state legislature is hostile to the will of the people.

When I watched the video I had the subtitles on. The automatic transcript is pretty good. "Test-time" which is used frequently gets translated as "Tesla" so watch out for that.

I get what you're saying about perception being reality and that ARC-AGI suggests beating it means AGI has been achieved.

In practice when I have seen ARC brought up, it has more nuance than any of the other benchmarks.

Unlike, Humanity's Last Exam, which is the most egregious example I have seen in naming and when it is referenced in terms of an LLMs capability.

Ah, this sheds light on the silence around the Windsurf acquisition:

OpenAI and Microsoft are at a standoff over the terms of the startup’s $3 billion acquisition of the coding startup Windsurf, the people said. Microsoft currently has access to all of OpenAI’s IP, according to their agreement. It offers its own AI coding product, GitHub Copilot, that competes with OpenAI. OpenAI doesn’t want Microsoft to have access to Windsurf’s intellectual property.

You are a perpetual motion machine. Truly prolific.

That is correct. Some historical context is much appreciated in this thread.

tl;dr on Section 174, Research & Experimentation costs went from being fully deductible in the year incurred to being deductible over a 5 year period.

Larger tax bills and a tightening on what roles/activities are deductible as R&E are likely what OP is pointing at with his comment.

To the best of my non-inside baseball research, Section 174 changes were simply one part of a package of revenue generating measures to offset the large tax cuts from the broader tax act they were a part of.

The changes came from The Tax Cuts & Jobs Act of 2017 that was introduced to the House of Representatives by Congressman Kevin Brady (R) Texas. The bill passed both houses of Congress along party lines. Then President Trump signed the bill into law. Section 174 changes did not take effect until 2021.

https://news.ycombinator.com/threads?id=heymijo&next=4332098...

FYI, OpenAI acquired Windsurf so valuation is not an issue.

I don’t know Varun (their founder/CEO) personally but I get highly competent vibes from him. I’d let my skeptical self lean on your optimistic take.

Yes, someone who truly wants greatness should be able to hear a poorly thought our argument, cognitively question the counter party to help them articulate exactly what they are thinking, or even articulate it better themself. Then you can weigh and measure competing arguments.

A high-functioning team is going to have at least one person who does this. For a perpetually high functioning team this is going to be second nature.

they told me their largest customers are for training, not inference

That is curious. Things are moving so quickly right now. I typed out a few speculative sentences then went ahead and asked an LLM.

Looks like Cerebras is responding to the market and pivoting towards a perceived strength of their product combined with the growth in inference, especially with the advent of reasoning models.

GPUs, very good for pretraining. Inefficient for inference.

Why?

For each new word a transformer generates it has to move the entire set of model weights from memory to compute units. For a 70 billion parameter model with 16-bit weights that requires moving approximately 140 gigabytes of data to generate just a single word.

GPUs have off-chip memory. That means a GPU has to push data across a chip - memory bridge for every single word it creates. This architectural choice, is an advantage for graphics processing where large amounts of data needs to be stored but not necessarily accessed as rapidly for every single computation. It's a liability in inference where quick and frequent data access is critical.

Listening to Andrew Feldman of Cerebras [0] is what helped me grok the differences. Caveat, he is a founder/CEO of a company that sells hardware for AI inference, so the guy is talking his book.

[0] https://www.youtube.com/watch?v=MW9vwF7TUI8&list=PLnJFlI3aIN...

You're right, but I think the general public is largely ignorant about calories and nutrition. Even basics like 4 cals per gram of protein and carbohydrate, 7 for alcohol, 9 for fats. Or even what a carbohydrate really is versus a "sugar", which people talk about as bad.

There's a sweet spot for an app that is inaccurate with a market that wants it but doesn't understand how inaccurate it is.

Kind of like how I could vibe code an app, get it to "work", think it's great and be ignorant of the many ways it will break or isn't working that a knowledgeable developer could.

And while this works for you, labor and employment attorneys use your non-standard application of the rules as a way to win lawsuits when brought against the company. Another way we end up with annoying, tedious, and distracting compliance (U.S. based take here).

I went into Project Hail Mary knowing nothing beyond it was from the author of The Martian. It may have been the most enjoyable reading experience I've ever had.

My guess is that they experienced something like this and needed larger sized normal shoes to accommodate the more natural foot shape.

https://www.reddit.com/r/BarefootRunning/comments/1ec5i5n/di...