ljoshua

To me, the common theme here in all of these hacks has been the company Irregular, which it looks like all the labs are using for sandboxing. But it looks like the sandbox may be a bit… lacking?

Yes, the models are smart when they find a way out, but their instructions are, in a way, deliberately open such as to be a case for misalignment in these cases anyway. It’s a capture-the-flag assignment in the Gemini case, and in the OpenAI cases they were broad instructions to best the reward function. As part of alignment studies, this is literally what you’re trying to observe and then work with. If Irregular’s sandbox had been a little more boxy, there wouldn’t be these issues.

I’m not saying we have zero problems here on the AI side, but I’d certainly be reviewing my contract with Irregular at this time if I were playing in the space. It would be just as interesting to learn more about their sandboxing techniques as it would the models in these particular scenarios.

Axo | Lead Founding Engineer | Sugar Land / Houston, TX (HYBRID) | Full-time | $120k–$160k + Founding Equity

Want to help out healthcare providers? Clinicians using our software are already saving hours a day and loving their job again. We are building Axo to make the software layer as invisible and ambient as possible, letting clinicians direct 100% of their attention to patients rather than screens.

Our first market is orthotics and prosthetics (O&P), the field designing braces and artificial limbs that restore human mobility and underserved by the tech sector. You’ll join as a founding engineer working directly with myself (former Google SWE) and my co-founder (one of the US's best known O&P clinicians), and small engineering team we want to grow.

You will own massive surfaces end-to-end, making foundational architectural decisions for our entire platform. We've got a massive pipeline ahead and need your help to build it. And we participate in humanitarian missions to expand prosthetic access in critical parts of the world like Ukraine and Sri Lanka.

- The Stack: C#/.NET, gRPC, SolidJS/TypeScript, and PostgreSQL, built with strict domain-driven design and rigorous testing.

- How we work: Flexibility! Love AI-assisted speed coding? Go for it. Prefer to write by hand because your mental model is faster than prompting? Perfect. We only care about delivering safely and quickly.

- Requirements: Senior-level depth with production systems, a strong habit of domain modeling, and a philosophy that defaults to simplicity. Strong UI/UX experience with an obsession for user simplicity is a must.

To apply, email me directly at jlyman@axoventures.co with a short story about a favorite technical challenge you solved end-to-end. Please put "HN" in the subject line! MUST be in Texas.

Upvoting if for nothing else than the intro paragraph to the repo. That was hilarious and so true.

Axo Ventures | Senior Software Engineer (Founding Team) | ONSITE: Houston, TX (Sugar Land) | Full-time | $100k–$140k + Founding Equity, adjustable

Clinical software is where a provider's time goes to die. We are building Axo to make the software layer as invisible and ambient as possible, letting clinicians direct 100% of their attention to patients rather than screens.

Our first market is orthotics and prosthetics (O&P), the field designing braces and artificial limbs that restore human mobility and underserved by the tech sector. We're building a foundation designed to eventually scale across thousands of providers and multiple medical specialties.

You’ll join as a founding engineer working directly with myself (Xoogler SWE) and my co-founder (one of the US's best known O&P clinicians), and a new engineer. You will own massive surfaces end-to-end, making the foundational architectural decisions for our entire platform. We have clinics using our suite and already saving hours a day, but we've got a massive pipeline ahead and need your help to build it. And we participate in humanitarian missions to expand prosthetic access in critical parts of the world like Ukraine.

- The Stack: C#/.NET, contract-first gRPC, SolidJS/TypeScript, and PostgreSQL, built with strict domain-driven design and rigorous testing.

- How we work: We have total flexibility on your workflow. Love AI-assisted speed coding? Go for it. Prefer to write by hand because your mental model is faster than prompting? Perfect. We only care about delivering safely and quickly.

- Requirements: Senior-level depth with production systems, a strong habit of domain modeling, and a philosophy that defaults to simplicity. Strong UI/UX experience with an obsession for user simplicity is a must.

To apply, email me directly at jlyman@axoventures.co with a short story about a favorite technical challenge you solved end-to-end. Please put "HN" in the subject line!

Liked this post a lot, well done! Definitely appreciated that it was a site that actual has a unique design, and isn't just another Medium article...

A book on a similar subject that I don't see mentioned very often but which I quite enjoyed as "Tough Things First" by Ray Zinn [0]. Not the most popular one, but really down to earth and approachable ideas. Kind of like PG's "do things that don't scale," just applied on a broader timescale.

[0] https://toughthingsfirst.com/book/

It really was/is a gem of a museum, very fun to visit and quite approachable. We went a couple times, once when they had some fly-ins that made it extra special.

Hopefully it can be preserved and continue it's life! There is hope: https://www.1940airterminal.org/news/texas-historical-commis...

> tl;dr: AI writes features, not architecture.

This. I definitely agree with this statement at this point in AI-assisted development. This gets at the "taste" factor that is still intrinsically human, especially in software engineering. If you can construct and guide the overall architecture of an application or system, AI can conceivably fill in the smaller feature bits, and do so well. But it must have a strong architecture and opinionated field in which to play.

I've had at least 256GB on my phones for the last couple of generations after having had to deal with storage issues beforehand, and it's been much nicer.

But I picked up a 16e for my son a few months ago, with 128GB, and yes, we're running into issues with storage space when it comes time to do an OS update. Between local music and photos storage, base storage, and the image for the new update, two or three times now we've had to delete stuff temporarily in order to get the update going. So I'm happy the new base is 256GB, at least that will probably last us a couple more generations before ~~640KB~~ 256GB is enough for everyone.

I’d love to see an article about designing for agents to operate safely inside a user-facing software system (as opposed to this article, which is about creating a system with an agent.

What does it look like to architect a system where agents can operate on behalf of users? What changes about the design of that system? Is this exposing an MCP server internally? An A2A framework? Certainly exposing internal APIs such that an agent can perform operations a user would normally do would be key. How do you safely limit what an agent can do, especially in the context of what a user may have the ability to do?

Anyway, some of those capabilities have been on my mind recently. If anyone’s read anything good in that vein I’d love some links!

Xoogler here (so I can’t help with any changes or feature requests now unfortunately) but yes, all of Google runs on Gmail. The amount of email I got as a software engineer there was crazy voluminous. I and most Googlers around me made heavy use of filters, labels, and all the text search operators available (see https://support.google.com/mail/answer/7190?hl=en&co=GENIE.P...). I also learned to operate Gmail purely via the keyboard with the built in keyboard shortcuts.

I’d occasionally have the frustration of not finding what I was looking for, but usually if I combined a search with at least one other operator (who it was from, what label it might have received, etc.) I almost always found what I was looking for pretty quickly.

And as for the signature image attachments thing, I think that’s actually an artifact of how the sender compiles the email, not Gmail. The “has:attachment” operator is one I use a lot and is usually quite reliable.

Hope that provides a little insight!

Hey @code_brian, would Tavus make the conversational audio model available outside of the PALs and video models? Seems like this could be a great use case for voice-only agents as well.

Just a plug for the book this content is derived from, Noam Wasserman's "The Founders Dilemmas." It lays out so many facets of startup decisions that deserve thought from the outset to prevent issues. It also strikes a good balance IMO between being based in statistics and research, and including anecdotes from actual experiences that bring the statistics full circle. I'd highly recommend it.

I’ve never been on the Spotify train, but with an all-Apple household, including HomePod Minis in multiple rooms, I’ve been stuck in iTunes/Apple Music land. We own our music, which is nice. And I dutifully pay the $24.99 per year for iTunes Match so that I can tell Siri what to play on HomePods, but I will be 0% surprised when they deprecate that service.

Anyone have a good non-Apple way of getting Siri to play songs from a personal music collection on HomePods? My kids use it most.

Just yesterday I paid my annual $24.99 iTunes Match subscription to keep my music library synced between my laptops, phones, and HomePods. It’s a beautiful thing, but it feels tenuous every time that renewal goes in. Will it be my last? I hope not!

There is just something about actually owning the music that appeals to me and my wife (and yes, we’re children of the original iTunes era, when you could load up your playlist and then click the cool nuclear-looking button in iTunes to burn it to a CD). It won’t last forever, but I’ll keep with it till it dies because it works.

Less a technical comment and more just a mind-blown comment, but I still can’t get over just how much data is compressed into and available in these downloadable models. Yesterday I was on a plane with no WiFi, but had gemma3:12b downloaded through Ollama. Was playing around with it and showing my kids, and we fired history questions at it, questions about recent video games, and some animal fact questions. It wasn’t perfect, but holy cow the breadth of information that is embedded in an 8.1 GB file is incredible! Lossy, sure, but a pretty amazing way of compressing all of human knowledge into something incredibly contained.

I was running into context window issues doing this. I could have gone in and split up the scanned book into chapters or something to get around this, and did that for a couple of subjects. But it wasn't too much work (and literally cost me pennies, like six of them) to get the pure text extract, and it's pretty easy to work with now. (Besides, which random dev doesn't love a little side challenge to explore new APIs at home every now and then? ;) )

It sounds like you may be speaking from experience, and if so, I respect that.

My kids have done both public schooling and now homeschooling. For a variety of personal reasons, public schooling was not going to be an option for a couple of them, so we're trying this out now and it has been successful. We are tightly integrated into a very active church group, and they have lots of social interactions on a regular basis there, as well as opportunities with other homeschooled kids around town.

It's definitely a balance, and there's no one silver bullet on either side of the fence, but the best any of us can do is actively strive for giving each child the best and most appropriate experiences for them.

Oh don’t worry, they make excellent use of their library cards. :)

NotebookLM audio overviews/podcasts have been an absolute boon for my homeschooled kids. They devour audiobooks and podcasts, and they love learning by listening to these first. Then when we come together for class, we discuss what was covered, and can spend time diving into specifics or doing activities based on the content. It’s super nice to have another option for a learning medium here.

To generate them, we’ve scanned the physical book pages, and then with a simple Python script fed the images into GCP’s Document AI to extract the text en-masse, and concatenated the results together into a text-only version of the chapter. Give that text to NotebookLM and run with it.

Source paper available at https://journals.aps.org/prx/abstract/10.1103/PhysRevX.15.02....

The challenge till this is widely supported (caniuse.com currently pegs it at 46% globally [1]) will be using this as a progressive enhancement that does not provide a worse or unusable experience for users with browsers not supporting it yet.

In other words, don’t include critical information or functionality in the new styling that isn’t available in the underlying plain select element! But such is always a good practice anyway.

Very nice to see this taking shape though! Should be a huge improvement over the div monster that custom select box replacements often are. :)

[1] https://caniuse.com/mdn-css_properties_appearance_base-selec...

Nice, great screenshots, thx!

"An unexpected error occurred. Please try again." Occurred after clicking the "Generate Questions" button. Console shows a 429 error. I'm sure your logging has picked it up, but just FYI.

Very cool! I was just looking for something like this last night to generate quizzes for my middle- and grade-school sure kids as supplement to their normal work. So far I’ve settled on NotebookLM.

One thing I’d love to see is some examples of what the generated quizzes look like/work like before I upload something. Right now, I have no idea what I’ll get before uploading something. Some examples or demos on the front page would be great.

Cool! Just a small note on naming: I assume you are going for a play of words on “infer,” but my brain (native English speaker but who also speaks French) immediately read the domain as “hell.ai,” because “enfer” is “hell” in French. ;)

Grid is fine, snow is melting, everything is business as usual. CenterPoint had 99.9% deliverability for the past 24 hours, and ERCOT has 14,781 MW in reserve power available (https://www.ercot.com/gridmktinfo/dashboards/gridconditions). Source: I live in Houston.

I know this was tongue in cheek, but c'mon, we can respect each other, right? :)

Unrelated to the discussion of JMAP, but I had the pleasure of helping host the mentioned Inbox Love conference, assisting Josh Baer and Jared Goralnick. That conference was such a fun one!

Targeted, focused conferences like Inbox Love, with 150ish or fewer attendees, are by far my favorite because you can actually get to know folks, ideas flow more easily, and everyone is focused on approximately the same thing. Much better than huge, multi-track conferences. We should host more of those as an industry.

I have two Kasa light strips (KL400) and anecdotally I’ve noticed that its performance degrades every other day or so to the point where it stops responding to change commands.

The fix? Blocking all inbound and outbound WAN (internet) traffic to it. Now works flawlessly, just like you think a light strip would. I only ever want to issue commands locally anyway, and why it should be talking to the broader internet in that case is beyond me.

Still not much, realistically 4096 bytes or less.

Browsers aren’t as much the issue as they’ve been in the past, but I’ve hit snags with proxies, old servers, etc.

I agree that the main takeaway is knowing when to switch. Having a mental model for this makes many future discussions and decisions much easier, because this seems to be a conundrum that comes up frequently (even for me inside of a large tech organization!).

Based on my experience, I'd suggest the pivot point occurs before even starting: it should pivot around who is building and maintaining the system. If you have the experience needed to quickly develop a solution in code, do it in code. If not, because this is a non-technical team without technical resources, do it in low code. Simple as that.

My team does Wednesday to Wednesday for many of the same reasons mentioned in the article, and it works great. We switch at 11am and hold a hand-off meeting at that time, and invite the whole team.

Hand-off meetings with the whole team work really well (in my opinion!) when you have a relatively small team--we have 9 FT teammates. Often someone else may have been delegated the page or bug that arose and can discuss how they handled it, or someone who wasn't involved may have insight for how to handle a situation better the next time. Since we're all going to be on rotation at least once during a quarter, it's great to know what happened in case a similar page pops up later.

Finally, we also fill out a running Doc before/during the meeting with links to the pages/bugs, along with short descriptions of how they were handled. This forms a great living memory of how to deal with incidents, and is also often the birthplace of new playbooks for handling new types of incidents.