silverlake
Joined 1,241 karma
- silverlakeI don’t know. But at least you’ve identified the real problem: lazy people generating trash code. AI isn’t bad, people are.
- Enough that stacked PRs are a thing. At my job people sometimes build large features on a branch for 6 months. Then it’s a massive PR and no one can review it.
- Hi, AI apologist here. This scenario is a problem with or without AI. You can’t drop a 13k line PR you don’t understand without prior discussion. There are many ways to use AI. Your scenario (keynote speech) is a bad way to use it. Instead, a PR where you understand every line, whether you or an AI wrote it, should be fine. It would be indistinguishable from human generated code.
AI is a tool like any other. I hire a carpenter who knows how to build furniture. Whether he uses a Japanese pullsaw or a CNC machine is irrelevant to me.
- I used ET but it requires a server process also. Some machines are too locked down to allow this. Wish there was a way to kick start the server on demand.
- It’s a glib analogy, but the goal remains the same. Today’s training sets are immense. Is there an architecture that can learn something with tiny training sets?
- I’m actually working on just this. What’s the smallest training data set required to learn tic-tac-toe? A 5yo doesn’t need much training to learn a new game, but a transformer needs millions of samples.
- Also, most of them are shockingly incompetent. It took years to assemble a list of quality service providers. I pay a little more but stuff works now.
- There is a mountain of code that needs to be written that can’t be due to costs. A project that needs 100 developers may be prohibitively expensive. But 10 10x developers would be within budget. Think health care, manufacturing, government, finance, etc.
- I tried to get Gemini CLI to update itself using the MCP settings for Claude. It went off the rails. I then fed it the link you provided and it correctly updates it's settings file. You might mention the settings.json file in the README.
- I’ve been talking to wealth management firms and am truly underwhelmed. AFAICT, they charge 0.5%-1.35% for therapy and the chance to put you in high fee products. Roboadvisors are a brilliant product for most people. In fact, a simple Boglehead portfolio is all you need. Most people have simple and similar risk profiles.
The feature I think would be useful is how to manage taxes. Roth conversions, selling the right lot, qualified dividends, tax loss harvesting, etc. A related feature would be generating income while minimizing taxes, i.e. Schwab’s Intelligent Income.
- Here’s the real rebuttal to my overconfidence in LLMs. Thanks for the link!
- No one is an expert on all the things. I use libraries and tools to take care of things that are less important. I use my brain for things that are important. LLMs are another tool, more flexible and capable than any other. So yes, grandpa goes to Best Buy because he’s running his legal practice and doesn’t need to be an expert on computers.
- You’re clinging to an old model of work. Today an LLM converted my docker compose infrastructure to Kubernetes, using operators and helm charts as needed. It did in 10 minutes what would take me several days to learn and cobble together a bad solution. I review every small update and correct it when needed. It is so much more productive. I’m driving a tractor while you are pulling an ox cart.
- Have you guys at Cline considered using LLMs to create summaries of files and complex functions? Rather than read a 500 line function, feed it a short comment on what the function is doing. I'd like to use a local LLM to create summaries at every level: function, file, directory. Then let the LLM use that to find the right code to read. This is basically how I navigate a large code base.
- No offense, but that video is brutally boring. Even at 1.5x speed I couldn’t get past 10 min. You should transcribe the audio and use an LLM to write a punchy sales pitch.
- If you look at the full table of NAEP scores, you see the differences are not that much. 8th grade reading is 245(WV)-266(MA). That's 20 point spread, where NAEP goes to 500 pts. A 4% difference between the best and worst state is insignificant. They all suck because they are at 50% of the scale.
- You’re getting downvoted because you’re making a few mistakes. 1) Energy Star is not a mandate, it’s a certificate if you want it. 25% of dishwashers are not ES at Home Depot. 2) Dishwashers are slow for a few reasons, a big one is gov’t stopped use of strong detergents. The new one needs time to dissolve foods. 3) “why solve X when Y is still a problem” is always a weak argument. 4) “markets will solve it” doesn’t always work because the individual cost of an energy guzzling appliance is a few extra dollars, but the collective cost is high.
The difference between appliances in 1970 vs now is immense. My dishwasher is so quiet we double check if it’s on. It uses less water than handwashing. Even the Chamber of Commerce (big business lobby) asked them to keep Energy Star.
- IIRC the effect size at 0.15 was narrowly for pre-teen girls on social media. Every other age, and all boys, were below 0.1 when looking at total screen time (i.e. games, youtube). Parents should check up on young girls, but most kids will be fine.
- China earns relatively little per iPhone/Mac. The complex components come from other countries like Taiwan, Korea and Japan. Apple earns a hefty profit for marketing savvy, iCloud services, and perhaps financing. China makes a pittance for screwing all the parts together. Where would you prefer to be in this product pipeline?
- I think BF16 and FP16 are 1979 TFPOPs, but FP8 is 2x faster at 3958 TFLOPs. So only 10% efficiency, down from 20%. That’s not good.