LLMs are not very good at performance issues unless you walk into it with a clear idea of the likely root cause and the bigger picture regarding actual hardware and desired customer experiences.

"Please make the code go faster"

vs

"I am noticing what appears to be contention between threads under workload A, B & C, but not with workload D".

These are completely different universes of capability and outcomes.

Even if an LLM can fight its way to the answer on its own, you can achieve a specific desired result much faster and with significantly lower risk if you are genuinely an expert.

I've seen a concrete example of this recently. I profiled the client's product "the hard way" and arrived at a change to a single line of code that would eliminate a mutex issue. One of the client's developers used the LLM and wound up with a change set that touched hundreds of files, but otherwise achieved approximately the same performance fix. The other developer even had my hint that it was a single file change and couldn't figure out how to do this despite prompting a leading edge model regarding this exact possibility over and over.

Taste and aesthetics apply to absolutely everything. Not just UI/UX design. Perhaps it is even more important that we care about the things that are invisible to the customer. It is certainly easier to forget about them or treat them like they don't matter as much.

If your product is handling natural language conversations from end customers, there is not much you can do to prevent the occasional PII leak without ruining the rest of the pie. ZDR is your best mitigation if you actually want the magical AI experience to work the way the investors hope it can.

PII can often become disclosed by way of many correlated factors that are not considered PII on their own. Even a perfect AI system cannot capture all of these relationships. You could probably locate where I live within a 20 mile radius if you spent enough time analyzing my HN comments over the years. Not one of these comments on their own would trigger a PII filter.

I actually don't mind the cost of living in my neighborhood going up. The $2500/yr HOA fee is a feature for me. I picked this location precisely because of it. I've lived in many places with virtually no maintenance overhead or economic friction. You may eventually learn that there are two sides to this coin. Neighbors who can afford and are willing to participate in ridiculously scaled housing markets also tend to take better care of their properties and local communities.

https://en.wikipedia.org/wiki/Nonlinear_junction_detector

As the value of a particular piece of software decreases, the value of reversing its specific implementation does as well.

Reverse engineering is about discovering specific methods or protocols. Not for porting entire implementations to new platforms or products. A lot of the specific methods and protocols have been sucked up into the LLM weights, so the need to go digging for these patterns is dramatically reduced.

The biggest application I see here is with digital archaeological work (the opposite of new things).

The mechanics of BFME are the important part, not the intellectual property of the films. The way the units move, the resource system and power points are what make it special. Gandalf and Lurtz are simply wallpaper on top of something that is already very competent.

Think of it as a more persuasive means of obtaining documentation. Not as a nefarious act.

There are still some shelf stable goods that you can buy without getting screwed. But it's exceedingly rare now. Peanut butter is a good example. 99% the inventory is JIF with palm oil and added sugar in a plastic container, but on the very top (or bottom) shelf you can often find a glass jar that contains only peanuts and salt. It took me a month to get used to peanut butter like this. The exercise of stirring it almost ruined it for me. I can't imagine going back to the other stuff now. I tried JIF again recently and it tastes like a product you'd buy from Home Depot to do a drywall job.

Why isn't this just one coherent agent loop with subtools/agents as appropriate? If these tasks are related in some way, having a single context would probably make it go much better.

The freewheeling messaging part is where the token bloat is coming from. I suspect that for some of us this is actually the point. I think it's a mostly form of entertainment to do things this way. The next logical step from Factorio gameplay.

Parallel agents remind me a lot about multi core compute. It's incredibly easy to take a single core product and make it run much worse across a lot of cores.

Light coming in from a window is generally filtered. Light being bright can be very deceptive. Even if the temperature is correct. Your eyes crave that deep infrared energy that LEDs and energy efficient windows actively reject.

I wish I could observe how some of us are using these tools.

I still struggle to spend $50 in tokens per month, and I exclusively use prepaid API tokens. This is in support of personal projects and two clients. There are billing cycles where I might spend upward of $400, but this is maybe once a year. This is offset by months like August wherein I spent $12 in tokens.

The other advantage with prepaid is that it handles the other direction much better. I don't even know what a quota limit feels like. Being blocked for hours is way more expensive to me and my clients than even $500/m. Losing an entire business day over this wouldn't work out.

I've managed to convince some others to try the same thing. $200/m flat fee is a pretty extreme constant expense if you can be more clever on average.

I think a lot of people are getting pushed around by FOMO effects into spending money on pointless subsidized tokens and have (valid) fears that if they don't maintain the same apparent economic leverage as their peers that they will be left behind. This isn't actually the case, much like lines of code are a really poor indicator for the quality or productivity over a codebase.

I make this "jump" about 20 times per day. I have iOS, MacOS and 2 windows machines at the moment. Very frequently I will RDP into the windows machines from the MacBook.

> goodbye Outlook-Teams-OneDrive-SharePoint-Excel-Word-PowerPoint-VBA

https://www.microsoft.com/en-us/microsoft-365/mac/microsoft-...

Maybe it used to be hard to do this and I just got used to it. I have made zero attempts to normalize anything.

It's a very compelling aspect of the problem.

If you can get a model under certain size thresholds, that means you can eliminate latency domains. For example, if the model is able to fit entirely inside L3, the latency of servicing requests drops by an order of magnitude (or better) compared with a model that resides in L3+DRAM.

I guarantee Micron & friends are not intentionally orchestrating their business such that they would suffer a massively reduced chance of yielding on a per-die basis. Unless someone is actually buying HBM devices, they are not going to be making them. These are not a commodity that can be speculatively manufactured in any economically rational way.

In a healthy grid with a diverse generation mix, batteries are almost certainly the perfect option for peak demand handling. The problem is that many grids are not healthy or do not have very diverse generation mixes.

You are effectively borrowing from yesterday (or last hour) to pay for today with batteries. Which is amazing, until it isn't. The Texas winter crisis comes to mind as an example where batteries would be regarded with intense derision. I think the current capacity market accreditation process is massively underrepresenting the tail risk of a black swan event. Four hours of battery storage should not be in the same room as a gas turbine when we are talking about capacity and multi-day emergencies.

Maybe these maintainers could simply cease doing all this free work? There are other ways to make a living. Perhaps even doing the same thing but with better negotiation skills. No one is forcing anyone to do this. The reality is that if 100% of the tzdb and SQLite teams disappeared today, Microsoft and Google would have these projects back under stable operation before the weekend.

We cannot expect a realistic economy to arbitrarily compensate people who originally did not seek compensation without a lot of really bad side effects. Does anyone on HN have any actual evidence that the maintainers of truly load bearing infrastructure (billions of devices+) are genuinely struggling economically?

What would prevent abuse of this kind of program?