chaisan

we moved our whole org off Vercel after that selfie Rauch put out. rotten company, overpriced product for what it is, sneaky practices. never looked back.

have examples?

nice. for which problem?

yess. loads of space for further exploration here. there is an attempt to keep things as general as possible in the expert.md file, but hard to mitigate overfitting fully. however, changing the seed will not get you much further with all else in the solver constant. unless you try a number of seed that exponentially scales with the size of the problem

sure. in the limit, everything is parameter tuning. with large enough NP-hard problems, the complexity of the search space is big enough that its infeasible to get to a better state by just tuning params in any reasonable amount of time.

wrt. token usage?

somewhat

and it would take an algo change to the solver to jump to the next local optimum

yeh. ofc. but on any problem larger than 40 variables, the gains from random restarts or initializations will quickly plateau

the sum of the weights of the unsatistied clauses. we want to reduce this number

as its from 2024 (MaxSAT was not held in 2025), its quite likely all the solvers are in the training data. so the interesting part here is the instances for which we actually got better costs that what is currently known (in the best-cost.csv) file.

its just comparing the cost of the best solution found to the best known cost we had before. O(N). why optimistic?

nice. EDA indeed one of the top applications of SAT

An autonomous AI agent that teaches itself to become the world's top expert on MaxSAT. Given weighted MaxSAT instances, it learns novel strategies, finds better solutions and iteratively refines its toolbox. No human guidance.

reminds me of this idea of Do What I Mean (DWIM) coined in the 60s by Warren Teitelman. more relevant now than ever

the scariest safety fuse isn’t in the model. it’s the gap between chair and keyboard

The workflow of starting with and iterating on a PRD is already taking off: https://x.com/daniel_mac8/status/1871372159748079893 https://skylarbpayne.com/posts/cursed-cursor

Ntropy (https://ntropy.com) | ONSITE, Full-time | Office in San Francisco and London, UK | SWE, ML | $130-200k

Ntropy is building domain-specific caching infrastructure for LLMs that makes it possible to reduce p99 latencies by 1000x and costs by 4-5 orders of magnitude for real-world, high-throughput tasks. Currently we are targeting the financial and banking domain and will expand over time. We are well funded by top investors (QED & Lakestar) and are launching an engineering office in SF this month. We are hiring for the following roles:

- Machine learning engineer (San Francisco, CA) - Full-stack / LLM engineer (San Francisco, CA) - (senior) Backend engineer (San Francisco, CA) - Front-end engineer (London, UK)

To apply, email [email protected]

nice. where is the the main part of the 4.4x boost on ARM coming from?

I keep running into issues where I update a selector which change the memory reference and causes performance issues. Have you found any good ways of avoiding that?

Great stuff! How do you measure frontend performance? Do you have any production metrics?

Ntropy | https://ntropy.com | data-QA, backend, full-stack | Remote | Full-time

Ntropy enables companies to build a new generation of products and services on top of financial transaction data.

We recently closed our series A and are hiring across data-QA, backend and full-stack roles to work on:

* Optimizing and extending high throughput ML-focused service * Developing new end-to-end financial insight products on top of our core technology * Stack: Python / PyTorch / Rust / lots of GPUs / Managed cloud services & k8s

Apply on https://ntropy.com/company/careers or email [email protected] if you have any questions!