bgirard

8 ms·

61.
by bgirard 1y ago
I think that's the real issue. If the LLM spends a lot of context investigating a bad solution and you redirect it, I notice it has trouble ignoring maybe 10K tokens of bad exploration context against my 10 line of 'No, don't
62.
by bgirard 1y ago
Did you attach the debugger and see what it was crashing on? From when I used to work on performance and reliability at Mozilla, these types of user specific crashes were often caused by faulty system library or anti-virus like software doi
63.
by bgirard 1y ago
FYI they are all used
64.
by bgirard 1y ago
We shouldn't tolerate verbal abuse to avoid upending the life of the abuser.
65.
by bgirard 1y ago
> hurts your feelings Let's not trivialize verbal abuse by calling it hurt feelings.
66.
by bgirard 1y ago
I'm most surprised that people used to spend 12.3 hours/week with friends. With work, errands, laundry, relationships, family I find it hard to believe. Plus anyone that's a parent to a young kid in that age range. That'
67.
by bgirard 1y ago
Do you have a citation? I don't believe that was ever the case when I worked on this.
68.
by bgirard 1y ago
I proposed will-change a decade ago when I used to work on the rendering stack at Mozilla and also tuned frontends. In some circumstances the heads up is totally required. For example if you're on a low spec mobile device with a relati
69.
by bgirard 1y ago
> after people getting orders for bacon on ice cream I let people, drunk teenagers trying to get their date to laugh mostly, add bacon to ice cream, flurries and shakes when I worked at McDonalds. It happened once or twice a year. It mad
70.
by bgirard 1y ago
That's really cool. Fascinating to think about all the problems that get missed due to poor or missing visualizations like this. I did a lot of work to improve the Mozilla build system a decade ago where I would have loved this tool. W
71.
by bgirard 1y ago
I agree. Opus is cost prohibitive for most longer coding tasks. The increase output doesn't justify the cost.
72.
by bgirard 1y ago
I'm playing with Claude Code to build an ASCII factorio-like. I first had it write code without much code supervision. It quickly added most of the core features you'd expect (save/load, options, debug, building, map generati
73.
by bgirard 1y ago
Mostly sonnet because of the usage limits.
74.
by bgirard 1y ago
That's neat. According to the tool I'm consuming ~300m tokens per day coding with a (retail?) cost of ~125$/day. The output of the model is definitely worth $100/mo to me.
75.
by bgirard 1y ago
YMMV I'm using the $100/mo max subscription and I hit the limit during a focused coding session where I'm giving it prompts non-stop. Unfortunately there's no easy tool to inspect usage. I started a project to parse the
76.
by bgirard 1y ago
> ~100b/s of UTF8 text out of a NLP black box is this big That's not a good measure. NP problem solutions are only a single bit, but they are much harder to solve than CS2 frames for large N. If it could solve any problem perfe
77.
by bgirard 1y ago
A combination of: 1) RUM, 2) Field tracing, 3) Local tracing with throttling (CPU, Network), 4) Focusing either on known problems, or if I'm on less certain territory I will do more vetting to make sure I'm chasing something real.
78.
by bgirard 1y ago
It's great. Much harder on Web because it's difficult to get rich information from the browser. This is why I contributed to the JS Self-Profiling API in the first place: https://developer.mozilla.org/en-US/do
79.
by bgirard 1y ago
> Identifying a representative usage scenario to optimize towards and then implementing that scenario in a microbenchmark test driver are both massively difficult to get right, and a "failure" in this regard, as the author foun
80.
by bgirard 1y ago
How does Bolt retain customers? Why would they burn customers to get 1-3 x €7 fees? If OP's account is representative then wouldn't all the customers switch to Uber and never look back? It doesn't make sense to me so it feels
81.
by bgirard 1y ago
I went through this after my divorce. The term we used in therapy was 'unhelpful thinking'. I've seen people sharing this tip on HN for dealing with tinnitus: Picture a volume equalizer or the FFT waveform and image yourself
82.
by bgirard 1y ago
The flip side of this is that I had a Ring doorbell in a home I lived in. Moved out and split up with my ex. Years later I installed a new Ring doorbell and I got a message from her shortly after installing it saying she was getting notific
83.
by bgirard 1y ago
I agree with that wholeheartedly. The problem is that people want to discuss today's news now. Explaining them this and not being able to engage in the conversation isn't a great way to connect with people unfortunately.
84.
by bgirard 1y ago
I worked on the rendering team at Mozilla during that time and contributed a lot to improving performance and reducing memory usage for B2G. I remember I was optimizing the experience for 256 MB devices, then one day we're told to star
85.
by bgirard 1y ago
> FFOS was doomed because the pixel explosion wasn't just stretching the GPUs of that time Actually this is the reason that I, at the time working at Mozilla on the rendering team and helping with Firefox OS, proposed will-change. T
86.
by bgirard 1y ago
Only if the LHC doesn't quire gold to operate. If you're using ICs and components that have some gold in them and they need maintenance, you consume more than you produce.
87.
by bgirard 1y ago
> People will buy it if it's a necessary item. What do you call necessary? What about non-necessary items? What about IP that improves manufacturing efficiency by 5% in some process? Better battery tech? Better UI? Better chips? Bet
88.
by bgirard 1y ago
I wouldn't call it logical for the US to lose all the IP protections they benefit from. But I doubt the international community would mind if suddenly American IP became free after the US withdrew itself from international trade.
89.
by bgirard 1y ago
> products we couldn’t manufacture ourselves even if we wanted to, since the vendor has well-deserved IP protections That's something I hadn't though about in the context of tariffs with interesting implications. Does following
90.
by bgirard 1y ago
Science working as intended. If claims are made, we can verify them and correctly prove or disprove them.

More ›