expected way more from a refresh tbh

same thing happened with crypto - the underlying technology is cool but the community is what makes it so hated

would love to know your actual day to day use case for what you built

i just got an m5 max with 128gb of ram specifically to run local llms

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=46715442

Points: 3

# Comments: 0

Will be using this a ton in the future

SimKit gives you a simple tick-based loop where agents can act, state evolves, and you can track everything deterministically with seeded randomness. It has built-in OpenTelemetry so you can see exactly what your agents are doing, and you can plug in any model or tool you want.

We started this because single prompt-response evals don’t really show how agents behave over time. With SimKit you can spin up multi-agent scenarios, watch state change step by step, and fairly compare models since runs are reproducible. No vendor lock-in, agent agnostic, and OTEL friendly.

It’s early but already useful for building testbeds, evals, benchmarks, and sandbox environments. Repo has examples and we’re adding more. Would love feedback or contributions (https://github.com/Fallomai/simkit)

Comments URL: https://news.ycombinator.com/item?id=45255861

Points: 4

# Comments: 0

It sort of makes sense, it's like the calculator all over again... we want to gauge how well a candidate actually understands what's happening, but it's also unrealistic to not let them use the tools they'd be using on the job.

After talking to a large number of companies about their recent hiring experiences, it seemed like their options were pretty limited. They'd either rely solely on in-person interviews, or they'd need to change how interviews were done.

We decided to build a platform that lets companies design coding interviews that incorporate AI into the mix. We provide two different types of interviews:

1. A web-based assessment that has an LLM on the left and a code editor on the right, and the candidate can interact with the LLM, explain their approach, and get guidance while coding if necessary.

2. A "work-trial"-based interview where the candidate has a set amount of time to complete the tasks that the interviewer has created. The candidate is allowed to use any resources at their disposal, and at the end of the interview has five minutes to upload the final code and their LLM chat export for review.

The company can decide what tasks and questions to add to both, that match what they're looking for. Also, we'd then allow the interviewer to use their discretion on whether the candidate compromised things like security, code style, and maintainability for shipping, as well as how well they vetted the AI's responses and asked for clarification and modifications.

Basically, the idea is to mimic how the candidate would actually perform on real-world tasks with the real-world tools they'd be using on the job. We'd also closely monitor the tasks and workflow of companies to ensure they're not taking advantage of candidates to get free work done, and that the assessments are actually based on tasks that have already been completed by their team.

Super interested in understanding what your thoughts on this kind of interviewing approach?

Link: https://fallom.dev

Comments URL: https://news.ycombinator.com/item?id=44226191

Points: 3

# Comments: 1

Basically as soon as you get paged in Slack by Jira or Pagerduty, we respond with:

- Past similar issues that might provide context - Relevant solutions from your company's confluence runbooks - POCs from other teams who may be involved in the issue - The ability to one-click create a war room - Instant setup. Just invite Pandem to the relevant on-call slack channels

We currently just launched our closed beta, and are accepting companies to test it out :)

Comments URL: https://news.ycombinator.com/item?id=40458675

Points: 1

# Comments: 0