tlrobinson

8 ms·

31.
by tlrobinson 3y ago
Bill Gurley’s recent talk is worth a watch https://youtu.be/F9cO3-MLHOM
32.
by tlrobinson 3y ago
See “Slaughterbots” short film: https://youtu.be/O-2tpwW0kmU
33.
by tlrobinson 3y ago
I'm curious if they actually mean "Zero trust" in the "perimeterless" sense ( https://en.wikipedia.org/wiki/Zero_trust_security_model ) or if they just mean their on-prem solution doesn't re
34.
by tlrobinson 3y ago
Maybe a dumb question, but why are media decoders, which are notoriously high risk, not well sandboxed?
35.
by tlrobinson 3y ago
> But 3 files that require updating where the code won't compile until you've added all three is way way less of a problem. I think this is only really possible in a relatively small subset of programming languages, even among
36.
by tlrobinson 3y ago
https://spectrum.ieee.org/amp/wall-street-tries-shortwave-ra...
37.
by tlrobinson 3y ago
Well, you could make both types in identical cups, put labels on the bottom of the cups, shuffle the cups, drink one, record your observations and take the label from the cup you drank and stick it to your notes.
38.
by tlrobinson 3y ago
You can also setup some reverse proxies to auth with SSO like Google. I use Traefik + https://github.com/thomseddon/traefik-forward-auth for personal projects, even on my local network.
39.
by tlrobinson 3y ago
const schema = fs.readFileSync(path.join(__dirname, "sentimentSchema.ts"), "utf8"); const translator = typechat.createJsonTranslator<SentimentResponse>(model, schema, "SentimentResponse");
40.
by tlrobinson 3y ago
I really like prefixed identifiers and strongly recommend using them from the beginning. One question: I’ve seen people use both integer IDs (as the primary key) and prefixed GUIDs (for APIs) on the same table. This seems confusing and wa
41.
by tlrobinson 3y ago
I agree with the addition of the checksum, however I’m curious: > either by accident or malice 1. if you don’t believe people hand type these then how else will they accidentally enter an invalid? I suppose copy/paste errors, or if
42.
by tlrobinson 3y ago
Jessica is one of my favorite people from the YC community, and I think that’s a commonly shared opinion!
43.
by tlrobinson 3y ago
> A study by the London School of Economics found pupils at phone-free schools performed better in GCSE exams, especially those in bottom 60% of KS2 tests https://www.theguardian.com/education/2015/may/15&#
44.
by tlrobinson 3y ago
I’m reminded of Louis CK’s rant “ Everything Is Amazing and Nobody Is Happy”: https://m.youtube.com/watch?v=kBLkX2VaQs4 Like, nothing like this was remotely possible a year ago. I’m sure we’ll hit a wall at some point, but
45.
by tlrobinson 3y ago
State charts are great for the right use cases. I used XState to help build a graphical editor (think Figma) with fairly complex interactions involving mouse and keyboard events, etc. I’m curious if others have found it worthwhile for more
46.
by tlrobinson 3y ago
If there’s one thing LLMs are pretty good at it’s summarizing content. Shouldn’t your website just have an “About” page with this information that humans can read too?
47.
by tlrobinson 3y ago
If I had a little more time I’d make a Twitter bot that used an LLM to extract unrelated / negative news from SEC filings and post it.
48.
by tlrobinson 3y ago
Yes! Having worked on both monoliths and microservice spaghetti this is how I'd do it. I wonder if at scale you could see additional benefits by further splitting the web tier by route?
49.
by tlrobinson 4y ago
Where are users reporting this?
50.
by tlrobinson 4y ago
I think yes, but more precisely the tokens were chosen to optimize training on a dataset that's biased to English content. I am curious how the token set affects quality of responses, ignoring the factors related to token count mention
51.
by tlrobinson 4y ago
I loved model rocketry as a kid. They even did it at my school (I won a contest in my class!) until a kid’s rocket landed on a neighboring house’s roof and burned it down.
52.
by tlrobinson 4y ago
> Alignment will be impossible. > we probably should start thinking more about defensive AI Isn’t this a paradox?
53.
by tlrobinson 4y ago
There’s also fine-tuning: https://platform.openai.com/docs/guides/fine-tuning
54.
by tlrobinson 4y ago
Or Postgres with pgvector https://github.com/pgvector/pgvector
55.
by tlrobinson 4y ago
You sir, are not Apple’s target market.
56.
by tlrobinson 4y ago
> The faster we build the future, the better. Past performance is no guarantee of future results.
57.
by tlrobinson 4y ago
I could also imagine a sort of two-tier approach, where the on-device model can handle the majority of queries, but recognize when it should pass the query on to a larger model running in the cloud.
58.
by tlrobinson 4y ago
Or perhaps a progressive summary, where the most recent messages are full fidelity, and older messages get “compressed” into a summary. You can also fine tune the model to incorporate larger amounts of data, but that may be more expensive (
59.
by tlrobinson 4y ago
> I want to be able to create a million lines of code per month by myself. That sounds terrible . Who is going to read all that code when you need to understand and modify it? The LLMs, I guess? I'd much rather be able to write 10,
60.
by tlrobinson 4y ago
IMHO it’s worth learning to properly type those cases, otherwise you never really know which parts of your application are actually being type checked. Personally I also find coming up with the correct types to be gratifying if they’re a li

More ›