ldjkfkdsjnv

o3 pro is really good, but the context is really constrained, so its hard to use and doesnt output enough. This makes it suitable for ideating on good abstractions, but cant really make broad sweeping changes. Grok will output a full file. If you use the o3 pro API, its actually great, but it gets really expensive.

I think its much better than opus. I would describe its code output as boring and to the point, no fluff. O3 Pro is better at abstraction, but grok heavy is better at bug hunting, and only doing exactly as needed. I swapped my openai pro license for grok, its good. Another big advantage is the context window size. Honestly I use these models all day long, and have felt that while sonnet 3.5 was ground breaking, but that anthropic is behind google, openai, and now xai.

For tools, I use repo prompt + grok website. Personally think claude code is overrated, and hand building the context by selecting the files is far better for complicated tasks

It happens all the time, I have seen it in NYC. Usually its an early stage thing, cofounder leaves after 1 year etc. Much harder to do with a complicated cap table. Investors I could name even suggest it

all that OCAML we only hire the smartest is often a veil for what is really a simpler operation that is borderline illegal. probably alot of employees dont even really understand the systems they work on

I know someone that made 10 million a year for a long time on wall street. They said, generally you can assume anyone making a large amount of money is a criminal. Any large deviations from the typical returns you would see in an asset class was suspicious

theres alot of shadow banning, up ranking and down ranking

This post is completely false

dont write any software until you have a product you are getting easy sales with, forget you even know how to code

ive been coding 5+ hours a day almost every day for 15 years. i think ai will replace 70% of SWE in the near future. not employement, but 70% of the current work done by engineers

all the jobs are being outsourced is why

people above and around you prefer if you stay within the range. over performing stresses other people out and causes conflict.

Yeah its called wipro/infosys/cognizant

if you are in a competitive market you will lose with this approach

This is all going to be solved by better models. Building agents is building for a world that doesn't quite exist yet, but probably will in a year or two. Building some big heuristic engine that strings together LLM calls (which is what this blog advocates for) is essentially a bet against progress in ai. I'm not taking that bet, and neither are any of the major players.

phd programs are largely immigration backdoors

The second you have a cofounder/business partner you need a lawyer

Dark truths are hidden in ancient dna

Theres two types of software, the ones no one uses, and the ones people complain about

I think this is the beginning of the end of early stage venture capital in b2b saas. Growth capital will still be there, but increasingly there will be no reason to raise. It will empower individuals with actual skill sets, rather than those with fancy schools on their resume

another article on xyz problem with LLMs, which will probably be solved by model advancements in 6/12 months.

theyve been automating S3 operations for 15 years. the critical AWS services are extremely mature, people have no clue

Inside view:

Amazon has a document writing culture, all of those documents will be written by AI. People have built careers on writing documents. Same with operations, its all about audit logs. Internally, there are MCPs that have already automated TPMs/PMs/Oncall/maintenance coding. Some orgs in AWS are 90% foreign, there is fear about losing visa status and going back, the automation is just beginning. Sonnet 4 felt like the first time MCPs could actually be used to automate work.

A region expansion scoping project in AWS that required detailed design and inspection of tens of code bases was done in a day, it would usually require two or three weeks of design work.

The automation is real, and the higher are ups are directly monitoring token usage in their org, and pushing senior engineers to increase Q/token usage metrics among low level engineers. Most orgs have a no backfill policy for engineers leaving, they are supplimenting staffing needs with indian contractors, the expectation being that fewer engineers will be needed in a years time.

as the models get better, IDEs will be seen as low level

They do poorly when you isolate some product, but if you look at their execution relative to their scale and number of employees, its exceptional. You have to see thats its all relative, google is a good big company.

"One of the most valuable companies on the planet is mismanaging its products"

see the issue with that assertion?

its valuable for the company, not for the line managers/product managers

yeah young business people dont want older engineers that can call them out around. they want other young clueless engineers to just build

Yeah ageism is a huge issue, even in early 30s, working as a regular programmer is met with suspicion and looked down upon.

Not only that, young engineers will be build whatever without questioning the value, business people prefer this. They dont want an engineer raising product questions

also, it is true that technology moves quickly, startups dont need experts in tech thats no longer bleeding edge. large old school companies still do, so the number of companies that are viable employers decreases

Word on the street is scale.ai knows which data sets/data approaches anthropic and openai used to do their RL reasoning training. Meta is paying for that know how/information.

Some of the worst behavior ive seen in my life has been done by devout catholics. who themselves cant see the consequences of their actions. and im not knocking religion, i just had to live alot of life to start being alarmed by outspoken moralists.