ramraj07

Herdr is not exactly obscure. I would argue your effort to follow ai hasn’t been that effective.

You tell this as if the Universities and departments where the research happened didn’t place themselves in ivory towers.

Pedigree mattered in academia. Gatekeeping was always rampant. So to a bystander like me this makes no difference.

Then point it to the Anthropic docs. Only place with reliable info. Look at their opus 5.5 prompting guide.

The best places where I get this info is from the ai engineer YouTube channel and Claude’s YouTube channel.

For effective harness engineering you need to also read the detailed prompting guide for each model. We are back at a place where every model revision introduces harness optimizations that need to happen.

You seem to be doing all the right things though. The tldr is treat your agent to be a precocious kid and the job of the harness is to helicopter parent it without the kid realizing it’s being parented.

The irony is that he’s still very very wrong in my opinion. He’s lamenting the end of math, but it’s not a stretch to think that math is just the beginning. Like the destruction of the buoy before the tsunami actually hits land. His current stance seems to be implicating that the blast radius is only till math or science when it’s just the first domino.

As far as I can tell, the AI team assembled in Apple has a fundamental instinct problem and I was hoping that the exodus earlier this year will fix it, but looks like it did not.

What would realistically be a ballpark expense for basic living (I assume you mean food water and power)

An ec2 or vps. But no sandbox can solve the problem that a true powerful harness would need access to the source of your information like slack and jira.

It’s not a kid, they’re 30 now looks like.

People made the exact same argument for years where Facebook stock stayed around 20 bucks post ipo. I did too and sold like an idiot!

This is not very international. Firstly DEI is apparently nongrata in the US now. On top of it GHC got a terrible reputation in recent years.

If you think about it that’s more likely than any other explanation.

Smart people can do dumb things. Their LinkedIn profile looks typical for an OpenAI engineer.

We also don’t know if this person did it under duress.

They made the judgement call to make private accounts free before the onslaught of AI. Now every coding agent is using GitHub as their file system backend and pushing it to limits no one has heard of.

If Microsoft had brains they would leverage it to become the de facto file system for all ai agents before cursor or anthropic figure it out.

How good is parakeet (assuming it’s Sota oss) compared to wisprflow? Wisprflow is magical ime

We don’t even know how acitaminophen works, thousands get their livers destroyed every year from it when alternatives exist, and we are yet to ban it. So how is all of the new reality any different?

Do you think the current clinical trial process has not been gamed close to the edge of uselessness ?

Do you know about the fiasco with the Alzheimer’s drug that has collectively cost humanity hundreds of billions?

I don’t see how an indecipherable AI based cure can be more misaligned.

Yes and I believe these two people who have been shown to be bastions of moral caliber.

It could be in the future but its not yet. Optogenetics has just merely been one more albeit powerful tool in the neuroscience toolkit and have generally not been impressed with how useful (or not) it has been..

Komodo Health | Principal Product Manager, AI Products | Full-Time | $213k-$288k + equity... | SF/NYC On-site / Remote US

Healthcare in the U.S. is a mess, and most of the data that could help is locked in places nobody can use. Komodo has mapped the patient journey across the country: diagnoses, treatments, providers, payers. Our customers are pharma companies, payers and health systems that want to act on that picture.

We're building the platform that puts it in their hands. There's an AI analyst that works across the whole map. There are apps that turn a population described in plain English into a full report in seconds, and tools that let analysts open, check and rerun the code behind every answer. Underneath all of it is a single query engine that every app and agent shares, so a new product is weeks of work instead of quarters.

Building has gotten cheap. Deciding what to build is the hard part now, and that's the job. We want product managers who:

  • Have shipped data, analytics or developer products that people use every day

  • Prototype things themselves with AI coding tools, and would rather show a working version than write a spec

  • Can sit with a customer, find the problem underneath the request, and say no to the rest

  • Write clearly and briefly
  Healthcare experience helps, but it isn't required. Curiosity about how the system actually works is.

Small team, real customers, and more leverage per person than I've seen anywhere.

If that sounds like you, email our head of product at [email protected] Or me at [email protected]

Please include the keyword from my HN profile in the subject of the email!

I have done something similar on a larger repo on claude code max and it didnt even fully exhaust the 5 hour limit.

DSPy never made full sense even back then in the somnet 3.5 era and it makes no sense today. Not a gut reaction but pure evals (we have had teams create the same agents using dspy, langgraph compete with pure agent loops and the agent loops come across as so much better by a wide margin). Its the people who insist on these procedural ideas that need to prove with data or money in revenue that these archaic ideas make any sense to consider.

Incidentally the teams and people who think dspy is great are also generally the worse teams that can never make a lovable product so take from that what you want.

Ive anecdotally heard that openai is far more chaotic, which includes not having a central infra team for example (or at least some teams not counting on depending on them). At least the previous hacks in openai were mainly due to bad infra architecture design.

Creating deep research multiagent systems without explicitly handing it a file system is a terrible idea. The only successful way we have found to reliably hand off context in multiagent systems is using a file system. Langchain even today doesnt make that default or even encourage it.

Theres more than words in our minds. Think harder are you absolutely sure? You REASON with words but your ideas dont form just from you reasoning. The ideas just seem to come out of nowhere to the part of your brain that then reasons around them.

Because many people (especially in HN) seem to reject allowing nature and capitalism take its course.

It (and most other systems) abstract the wrong concepts when you want to create an agent. They promote what was likely never a great strategy but even parsimoniously what are strategies that were good ideas months or a year back.

For example how langchain or whatever else handles subagents and deep research is laughable even today.

So whats the recommendation? Just use pydantic and code everything up yourself. Maybe strands I dont know.

By all indications Chris Cornell likely tried to paint some poignantly depressing mindset with that lyric, so its doubly ironic the universe is choke full of it. Or at least was.

The largest supermassive blackholes have an average density less than earth's atm as well.