Puspesh

Hard-to-please product engineer, Day dreamer, 2x Tech Entrepreneur (1x Exit). Into books📕 + podcasts 📻 and learning something new🧑‍💻

[Derek Thompson @DKThomp]

This has quietly been a miracle month in medicine.

In the last 5 weeks we’ve got news on:

- retatrutide, the triple agonist GLP-1 from Lilly, basically melting fat and body-wide inflammation at record levels
- RevMed’s new pancreatic cancer drug showing unprecedented abilities to extend life
- small trial of a one-and-done PCSK9 gene editing therapy for slashing LDL cholesterol
- Mayo’s AI-assisted radiology showing vastly improved cancer detection
- this new therapy for metastatic solid tumors

This stuff is at varying levels of evidence. Retatrutide is ~100% on its way, other stuff needs more clinical trial data. But put it together and we’re maybe on the verge of majorly reducing the mortality of heart disease and cancer, the two leading causes of death in America.

[Puspesh @puspesh]

Transfered to ICSE or NCERT? For replicating the amazing transformation?

[Ravi Handa @ravihanda]

Transferred!!!

Not sacked. Not even suspended.

Someone should lose their job over this. Ideally, the education minister but I would be happy if it is a senior govt official.

[Puspesh @puspesh]

I think he is talking about agents.

Yeah that’s how we decide whether to opus or go glm/kimi

[Sumanth Raman @sumanthraman]

All those giving gyaan here that I should have asked his expectation before evaluation, is this the way it is done nowadays? So we first ask the resource what he expects and then if we can afford it we evaluate? Seriously?

[Puspesh @puspesh]

@zodchiii Built open source version of this for personal use. Use with your existing Claude code subscription or APIs, all local with skills and customizable team.

https://t.co/dI3jLKkPaT

[Puspesh @puspesh]

AI is just a tool

[Mitchell Hashimoto @mitchellh]

I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem.

As an experiment, I rewrote the Ghostty core render state in Go, with access to identically laid out data structures as Ghostty and the exact same validation tests. I made a purposely naive renderer (simple, correct, but slow). 88ms per frame with 150,000 allocations (horrendous, lol)!

I then kickstarted a Ralph loop to bring the frame times down. I told it it can't modify input data structures or the public API or tests (they're correct), but it can do anything else it wants. It got to work.

It has worked for about 4 hours. I've spent around $350 on this experiment so far. The results?

88ms => 1.5ms
150K allocs => ~500 allocs

Incredible right? Nope.

My hand-written renderer I ported has frame times (same benchmark) of ~20us (0.020ms) and 0 allocations in the update path.

This is the problem with psychosis and lacking systems understanding. If you don't understand the system, you're going to accept that this is an incredible result. If you understand the system, you'll see better solutions immediately and can do roughly 75x better on throughput.

The people who blindly trust agent output are in the former camp. They're sheeple, overdrinking from a fountain of mediocrity.

Standard disclaimer: I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.

[Mitchell Hashimoto @mitchellh]

I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem.

As an experiment, I rewrote the Ghostty core render state in Go, with access to identically laid out data structures as Ghostty and the exact same validation tests. I made a purposely naive renderer (simple, correct, but slow). 88ms per frame with 150,000 allocations (horrendous, lol)!

I then kickstarted a Ralph loop to bring the frame times down. I told it it can't modify input data structures or the public API or tests (they're correct), but it can do anything else it wants. It got to work.

It has worked for about 4 hours. I've spent around $350 on this experiment so far. The results?

88ms => 1.5ms
150K allocs => ~500 allocs

Incredible right? Nope.

My hand-written renderer I ported has frame times (same benchmark) of ~20us (0.020ms) and 0 allocations in the update path.

This is the problem with psychosis and lacking systems understanding. If you don't understand the system, you're going to accept that this is an incredible result. If you understand the system, you'll see better solutions immediately and can do roughly 75x better on throughput.

The people who blindly trust agent output are in the former camp. They're sheeple, overdrinking from a fountain of mediocrity.

Standard disclaimer: I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.

[Puspesh @puspesh]

@danshipper @nbaschez Would love you folks to try and give feedback if any. (Free to sign up and start) @danshipper @nbaschez

[Puspesh @puspesh]

@danshipper @nbaschez Built a tool to even give Claude code/codex to the candidates and see “how” they build. (Recording, evaluation, note taking, proctoring incl)

https://t.co/J1ZQi71JWf

[Puspesh @puspesh]

@banglani Esp after a waffle!

[Puspesh @puspesh]

Every agent needs a human! @danshipper deep insight this

[Dan Shipper 📧 @danshipper]

dropping something heavy soon

[Puspesh @puspesh]

Swaad, Pan pasand, eclairs.

[Aravind @aravind]

What is your favorite Indian toffee?

[Puspesh @puspesh]

Obviously there was. It was Andrej interviewing them on why should he join them. Are they worthy?

[himanshu @himanshustwts]

i am genuinely curious about interview process (if any)

[Puspesh @puspesh]

Reserved instances reborn!

[OpenAI @OpenAI]

Introducing OpenAI Guaranteed Capacity: a new offering that enables customers to guarantee long-term access to OpenAI compute.

We’ve made long-term investments in infrastructure, partnerships, and capacity planning to help customers scale reliably.

Now, Guaranteed Capacity helps customers plan ahead for critical workloads in a compute-constrained world.

https://t.co/TN4OkZr2Uo

[Puspesh @puspesh]

@_swanand Ouch.

[Puspesh @puspesh]

@AjeyGore Openclaw is too technical for even tech people

[Puspesh @puspesh]

Not sure whats the efficiency gain here? Cleaning effort increases, egg breaking effort decreases.

[MERICA MEMED @Mericamemed]

A KEUREGG for eggs in the morning. Lazy but low key want it.

[geoff @GeoffreyHuntley]

my talk from @aiDotEngineer singapore is now up.

a. idk where this goes, no-one does.
b. i want you to disagree or agree but mainly reflect.
c. enjoy the animal style of a talk delivered whilst wearing an in-and-out hat.

https://t.co/NWMe1IEnBy

[Puspesh @puspesh]

My working model with agents - define tasks and then get out of way. If you can do it "async" .. do not waste time in doing "sync".

Running multiple teams (dev, content, research ..) with https://t.co/dI3jLKkPaT

[Vivian Balakrishnan @VivianBala]

Built a personal AI agent - why I did that and what it means. https://t.co/lo8SLUMD9s

[Puspesh @puspesh]

@nejatian Very cool. After applying, when you want to see "how" they used AI, we recommend using @fairground_work to conduct AI-native technical rounds