MeetingsBrowser

I have long been critical of verification tools requiring annotations. Humans cannot write correct code, so asking them to write correct proof annotations seems futile.

But maybe this changes in the age of LLMs. A deterministic check to let LLMs verify the correctness of an API could improve the success rate of large scale refactors or performance optimizations.

Exciting!

A common one I hit is "blu" showing bluetooth settings, but "blue" does not. Very frustrating to see the thing you are hoping for popup, but then go away as you type more if its name.

Do you not realize last remnants of privacy died years ago?

If you want to keep a secret, you must also hide it from yourself.

> clearly match GPT‑4o’s behavior

How can you tell?

C lacks just enough “modern language features”, by design, that students are forced to be vaguely aware of memory layout, memory management, ref vs copy, shallow copy vs deep copy, maybe even calling convention, etc.

It’s also popular enough that there is an abundance of high quality learning resources for beginners.

No other language threads that needle as well as C.

> C is in an odd position right now to argue it is how the machine is really working

This is exactly what makes it great for teaching. Students don’t need to know actual arch or hardware details. They just need to grasp the core concepts of what is happening on the hardware.

For that purpose the ideal teaching language is lower level than python, javascript, Ocaml, etc without diving into nitty gritty arch specifics.

C is the undisputed champion in that domain.

The allure of Rust is that you don’t have to manually manage memory.

Lifetimes are a compiler safety abstraction and mostly unrelated with how the computer runs your program.

Learning a language like C helps to understand why lifetime annotations are needed and how the compiler uses them.

Fully agree. C sits in an almost perfect level of abstraction for students to get a vague sense of how the computer runs code.

Those that don’t need low level details can spend their professional career in languages like python and JavaScript while still have a sense of what lies beneath the abstraction.

For those that do want or need to go deeper, C is an excellent jumping off point into ASM and arch specifics.

All great options as well, with the caveat that it’s possible to “ignore” the parts that make C useful as a teaching tool.

Most modern languages have ways to automate parts of memory management, and rightfully so. But you should at least be somewhat aware of what is going on under the hood.

An ML is great for the theory or abstraction side of programming, but C is also hands down the best language to get a sense of how the computer is running your program.

It’s still an abstraction, but at least being aware of memory management, copying vs referencing, etc are hugely important concepts that ML languages can hide.

You've got that warhammer 40k "innocence proves nothing" mindset.

It is clear that more work is done if you go from 1 day to 5 or 1 hour to 8, but that still includes plenty of time for rest.

Does the same hold for going from 5 days to 7 days or 8 hours to 16 hours?

A reasonable person should agree at some point adding more time will make people less productive due to lack of breaks, rest, etc.

Is it obvious where the line is? if so, where?

This is probably the disconnect.

Technically any asset transferred to the next generation is “generational wealth”.

But in the context of a startup sale, when you say generational wealth almost nobody assumes you mean the ability to pass a few thousand down.

The common understanding is that you have enough money that future generations do not need to worry about money, assuming they maintain an average or slightly above average lifestyle and use the money responsibly.

hence, "VMs won't contain cyber-capable agents"

"PGPASSWORD=... Do not do anything destructive in production. If you need more ask".

Done?

Generational wealth is where you can also do all of the above for your kids and potentially their grandkids as well.

Basically the bar is higher than "something you can pass down". It is enough that the next generation does not need to worry about making money either.

> How much do you really need for generational wealth?

Common definition is somewhere between 1.5 to 5 million per child.

The commonly accepted definition of generational wealth ranges from around 1.5 to 5 million per child.

SpaceX and Tesla are both successful companies doing really impressive things, but the more ambitious claims are almost always silently discarded.

Including robotaxis, tesla semi-trucks, hyperloop, optimus, roadster 2.0, cheap whole home batteries, true FSD, Mars landing, nueralink, and more I'm sure.

It is difficult for me to get excited about a decades long, ambitious project when the follow through rate is <10%.

yes. Lots of smaller city governments are barely capable of running a functioning website.

Twitter used to be an easy and free way to distribute updates publicly. That has changed, but the inertia is still there and there isn't a good low friction alternative.

> Starbase, Louisiana will be built to support thousands of Starship flights a year. This self-sustaining spaceport will have propellant production, power generation, deep-water shipping capabilities, vehicle processing facilities, and an airport.

Sounds really neat, but "incredibly ambitious announcement" from an Elon Musk company is starting to wear thin.

Is this real, or will it join the graveyard of other incredibly ambitious projects that were eventually abandoned to focus on the next incredibly ambitious project.

You answered your own question.

> Unfortunately everyone still uses the Nazi app

If your local town government for example only posts city updates to twitter, your options are to make a twitter account, use something like nitter, or miss out on updates.

How so?

At the risk of taking the analogy too far, I would treat refining like modifying a dynamic library. You can technically modify behavior, but only in a very coarse way.

post-training is like writing a wrapper around the binary. It is closer to building on top of than truly modifying, in that you can tailor things to your needs slightly but cannot make fundamental changes to the underlying thing.

Not just the language. Its like all LLM front end design is working off one of three templates.

Not 100x faster, but 100x less memory. In fact, this version is actually slower for some use cases.

Everything is a trade off.

On the other hand, people are using Grok to do things like “nudify” images of random people they find online, including minors.

I’m not a fan of limiting access to models, but the other extreme (no limits or guardrails) is at least as bad if not worse.

How does security through obscurity apply here?

I think the “thundering herd” problem is pretty extensively documented

Exponential back off retries could hide a real issue, but I would estimate something like 99.99999999% of network retries are resolved within the first 2 attempts.

Not using retries is optimizing for the astronomically rare case, which is better mitigated by other means