shishy

Do you think it's impossible for someone to have a 25K line PR / diff but not know what's going on in the system?

I ask because I've pushed some of this to the limit in my own testing and when I get the right building blocks / primitives in place in the repo, I'm able to get plans written in a way where I don't just describe what I want but specifically mention outcomes, verifications, constraints, etc. that utlimately describe the shape of my system and my intent.

My plan is my understanding of my system and the changes I'm making, and the execution of that plan (e.g. with subagents), verified against my spec, is the code itself.

With the correct building blocks, I've found that frontier models can write code in fairly predictable ways that lets me maintain my understanding and confidence in pretty large PRs.

I think perhaps the quoted part of your post seems to imply it's not possible / isn't phrased with that nuance?

> In fact, the most dangerous model may be one that is trained in secret and handed only to the People’s Liberation Army for use in drones and the Ministry of State Security for surveillance and repression.

Aren't Anthropic models used in project maven: https://en.wikipedia.org/wiki/Project_Maven ?

yeah. they could have called it Medjourney.

This must be the Opus 4.7 constitution

i used to watch this on repeat when i was a kid. it makes for a great christmas movie these days :)

Hi - the opportunity space page is meant to be broad; a more specific and targeted programme thesis within that is going to come out in the coming months. I should mention that it's not just a call for technical solutions because, as you note, a lot of this is deeply rooted in social systems and requires design solutions (i.e. it's not just building more tech). But that is good feedback!

The idea that the future is architected by our choices (or lack of it) is the crux of one of the opportunity spaces at the Advanced Research and Invention Agency (ARIA) in the UK.

Worth getting on your radar if this stuff is of interesting: https://aria.org.uk/opportunity-spaces/collective-flourishin...

(full disclosure: I'm working with the programme director on helping define the funding programme, so if you're working on related problems, by all means share your thoughts on the site or by reaching out!)

People are probably copy pasting already without that disclosure :(

> It was easy to swallow that failure. I made an honest mistake; I expect companies to reject candidates who make honest mistakes during interviews.

what a ridiculous statement

Yes well bubbles are a core part of the innovation process (new tech being useful doesn't imply a lack of bubbles), see e.g."Technological Revolutions and Financial Capital" by Carlota Perez https://en.wikipedia.org/wiki/Technological_Revolutions_and_...

I agree with you but there's something ironic about seeing that comment here, especially considering how many jobs tech has replaced in the last few decades without people having the time to retrain.

They probably meant a "once in a generation" startup like a unicorn

Yes, it depends on what you're doing; for general paper discovery / search tasks, title abstract can be enough (which is also why Springer and Elsevier have been pulling even their abstracts from sources like OpenAlex).

But for something like that you need full texts to look into results sections. I'm very curious how you're dealing with information contained in tables, or if you're dealing with snippets of text from the full-text alone. Have you poked around Elicit yet?

grobid is a wonderful resource, patrice did an awesome job (I used it at my previous job at scite.ai)

I always say that Iowa City is the Paris of the Midwest ;). Except when there's a home game.

The art culture is incredible. Two of our favorite artists spent a considerable amount of time there, so we made a few trips when researching trying to find pieces.

There is craft in business, in product, and in engineering.

A lot of these discussions focus on craft in engineering and there's lots of merit there regarding AI tools and how they change that process, but I've found that folks who enjoy both the product side of things and the engineering side of things are thriving while those who were very engineering focused understandably feel apprehensive.

I will say, in my day job, which is often at startups, I have to focus more on the business / product side just given the phase of the company. So, I get joy from engineering craft in side projects or other things I work on in my own time to scratch the itch.

You think it's a two way door, but the one behind you will close shut and you'll realize it was a one way door once someone says: "but 48 hours wasn't enough to evaluate it" ;)

> If there's a structural ratchet mechanism, where outsourcing a particular kind of manufacturing work means the US can never do it effectively again, doesn't every tick of the ratchet pose significant risks?

Yes, and in an earlier post Ben made that argument and said that the broader conversation on tariffs has been straw manning and wasn't trying to understand where they were coming from. He did more or less agree that the post-ww2 and post-nixon economic setup created this and articulated why it's problematic.

His thesis is just that this approach is wrong, and that it's hard to unpack the reasons because you don't know what to trust: is what the admin says a PR problem, a bad argument, directional or literal, etc.

Here's the older post (the one I shared in this thread was a continuation): https://stratechery.com/2025/trade-tariffs-and-tech/

Thanks all for reaching out; we found someone! I love HN.

If you see this way later and are interested, of course feel free to write me a note. You never know :)

Yep. It was a little shocking when I learned that about HCB a while ago. Totally agree.

The interesting part of her work is that unlike other photographers who not only took photos but had the chance to edit them, someone else discovered a bunch of negatives she left behind. So others came in later to turn them into what you see.

Anyone who has shot photos (film or digital) knows how much effort goes into the editing process.

I love her photographs and her story, I just find it fascinating to think about this detail, especially when contextualizing her with other artists.

Portal Stargaze | Fractional Engineer | Part time, 10-20 hours/week, $100-$150/hr (potentially full-time later) | Remote ----------------------------------------------------------------------------------------------------------------------------------------

Portal Innovations invests in ecosystems to support life sciences entrepreneurs in different geographies in the US. We use Stargaze, our AI-toolkit for identifying and sourcing innovation to make strategic decisions around:

- Finding rising star researchers to support

- Understanding where science is headed

- Forming high-impact relationships with researchers, founders, investors, and pharmaceutical companies

As we continue to build out this "frontend for biotech innovation", we need a seasoned engineer to help us accelerate our roadmap for 2025.

Here's a snapshot of what you'll work on with me:

- Build ETL pipelines (bringing in data around publications, patents, etc.). We've made good progress here, but need someone who can get it over the finish line.

- Build an internal data infrastructure that allows us to iterate on data science experiments to create new Innovation Biomarkers

- Work with our team to evaluate and benchmark different models and pipelines to classify innovators, resolve entities across datasets, generate high quality matches for innovation search, and more

- Occasionally work across the stack (React / Next, Python, Postgres) if you're interested in branching out

You're a great fit if you:

- Have experience with Python, Postgres, and general ETL pipelines / infrastructure (AWS, Modal, Airflow)

- Take your craft seriously and own your work

- Aren't afraid of getting into the weeds

- Work well asynchronously but appreciate the occasional meeting to iron out any blockers

Bonus points if you:

- Enjoy reading about machine learning & data science or have trained and deployed models in the past

- Have an interest in the biotech innovation space

Drop me a note at ashish.uppala@portalinnovations.com if you're interested, or know anyone that'd be a good fit :)

I figured those contracts with an intermediary would extend to anyone they re-sell to, or prohibit them from re-selling...

It took a few tries to get the verification code, probably overloaded

Thanks! I used to subscribe but it's been a while.

Bummer this one is for paid subs only :(

I mean I think the OP is referring to sales call for differential pricing. Any mature product would have product team looking at active accounts (even if enterprise sign up was self-service) and scheduling calls to understand needs and drive product improvements. There's never a substitute for that for the reasons you said.

There are other preprint servers. But to your question, there are centralized indices that track all papers.

DOI is the primary identifier and preprints are also issuing them now.

Crossref has papers by DOI. OpenAlex and SemanticScholar also have records, with different id types supported (doi, pmid, etc).

paperetl is cool, saving that for later, nice! did something similar in-house with grobid in the past (great project by patrice).

I enjoy seeing projects like this!

If you expand beyond arxiv, keep in mind since coverage matters for lit reviews, unfortunately the big publishers (Elsevier and Springer) are forcing other indices like OpenAlex, etc. to remove abstracts so they're harder to get.

Have you checked out other tools like undermind.ai, scite.ai, and elicit.org?

You might consider what else a dedicated product workflow for lit reviews includes besides search

(used to work at scite.ai)