HNHacker News
TopNewBestAskShowJobs

camjw

641 karma · joined October 20, 2018

Cracking on.
submissionscomments
camjw··on Show HN: GrapeGPT, AI sommelier/wine nerd
Ah thanks - there was an issue where I was sending absolutely massive payloads. Maybe have another go, should be quicker now. Made this in like six hours though so gimme a pass
camjw··on Show HN: June.so for LLMs
Thank you! Love your product!
camjw··on OpenAI builds first chip with Broadcom and TSMC, scales back foundry ambition
Surely this “logic” applies to every positive amount of dollars. Why not 700T
camjw··on TypedDicts are better than you think
Sorry this is probably a stupid question - how do you make your CI hard fail if you don't have types? This sounds like the missing piece for me, someone who also prefers just to crack on without types and then add them later.
camjw··on Ask HN: What's the "best" book you've ever read?
Not sure which one I would pick but the Culture series by Iain M Banks probably had the most impact on me.
camjw··on Ask HN: What happens to ".io" TLD after UK gives back the Chagos Islands?
"That part of the world" - christ mate its just a military base
camjw··on "Our view is that PearAI should start over from scratch"
I think the current title (i.e. including the word "Demands") is a bit more aggressive than necessary. IMO the Continue post is very sensible and even handed - the relevant sentence is "Our view is that PearAI should start over from scratch".
camjw··on Y Combinator Traded Prestige for Growth
I've used meticulous.ai!
camjw··on OpenAI to become for-profit company
I know this is just a short history but I think it is inaccurate to say "2015: Ilya Sutskever founds OpenAI." I get that we all want to know what he saw etc and he's clearly one of the smartest people in the world but he didn't found OpenAI by himself. Nor was it his idea to?
camjw··on Ask HN: Kotlin SpringBoot vs. Python Django for Min Viable Product
Use whichever one you know best already
camjw··on AI solves International Math Olympiad problems at silver medal level
Seems fair - I strongly think that "certain formalized problems" is very very far away from doing actual maths! And that actual maths is sometimes much less about solving known conjectures and much more about developing new theories.
camjw··on AI solves International Math Olympiad problems at silver medal level
I think that is unlikely to be the case - the classic example of a proof that human's "can't understand" is the Four Colour Theorem, but thats because the proof is a reduction to like 100000 special cases which are checked by computer.

To what extent is the proof of Fermat's Last Theorem "incomprehensible to humans" because only like a dozen people on the planet could truly understand it - I don't know.

The point of new proofs is really to learn new things about mathematics, and I'm sure we would learn something from a proof of Goldbach's conjecture.

Finally if it's not peer reviewed then its not a real proof eh.

camjw··on AI solves International Math Olympiad problems at silver medal level
> Theorem proving is a single-player game with an insanely big search space, I always thouht it would be solved long before AGI.

This seems so weird to me - AGI is undefined as a term imo but why would you expect "producing something generally intelligent" (i.e. median human level intelligence) to be significantly harder than "this thing is better than Terrence Tao at maths"?

camjw··on AI solves International Math Olympiad problems at silver medal level
Sure but if an AI can prove e.g the Goldbach conjecture then that is a bfd.
camjw··on My Python code is a neural network
For your interest, Taylor Series are not universal function approximators - the Taylor Series around 0 for

f(x) = e^(-1/x^2) if x != 0 else 0

is identically zero (all partial derivatives are 0 at 0) but the function is clearly not identically zero. So the radius of convergence for this Taylor series is infinite but it only equals the approximated function at one point.

I'm sure there are some conditions you can put on f to make the Taylor Series a UFA but it's been quite a while since I did any real analysis so I have forgotten!

Doesn't detract from the overall point though that there are UFAs that are not neural nets. I should say that I don't know what the precise definition of a UFA really is, but I assume you have to have more than equality at one point.

camjw··on Self-Serve Dashboards
Why is it that $BILLIONS is spent on this problem then, if it doesn't exist. I get that, yeah, management and execs only care about 3 KPIs but presumably people lower than them in the chain care about improving those KPIs and digging into the data helps you do that?

Or is this not true?

camjw··on Self-Serve Dashboards
How is this any different (text-to-sql vs text-to-semantic-query). Isn't this just comparing text-to-sql to text-to-slightly-simpler-sql?
camjw··on Self-Serve Dashboards
This feels like one of those spaces where everyone says LLMs and AI are just going to make it a solved problem, but no-one in the comments here is even mentioning it - maybe not a good problem for LLMs after all?
camjw··on Ask HN: Most successful example using LLMs in daily work/life?
GitHub copilot and nothing else comes close tbh.
camjw··on The Wait Equation and AI Investment
At least you concede then that Google invested in the hardware.
camjw··on Show HN: Built a Tool to Automate Tedious Tasks with AI, Without Coding
I've been using the alpha version of this to help recruiting engineering candidates, has been really useful so far (just saves me scrolling through LinkedIn etc). Basically I give it my notion of candidates with their LinkedIn and it generates some useful copy that I can include in an email, which it then emails for me.

Super helpful and Lorenzo has been very responsive - would recommend.

camjw··on HashiCorp switching to BSL shows a need for open charter companies
Freeloaders as in people who use e.g. terraform, react, postgres but never contribute to it?
camjw··on Show HN: Learn a language quickly by practising speaking with AI
This is really super cool. I wonder if you could use elevenlabs or something similar to make the voice more legit because it feels at the moment like I’m speaking to a satnav (in French).

I also got fluency score None which, let’s be real, is pretty accurate. But I feel like I need a bit more to help me learn.

The microphone interaction is really nice.

camjw··on Show HN: Superflows - Open source toolkit to build AI assistants for SaaS
This looks so useful! I've worked with so many legacy tools that have awful UX and only some random graybeard knows how to use, I'd love to add a copilot to some of those tools so that I (and my non-technical coworkers) don't have to figure out how to use them.

I'm thinking about random ERP tools, internal dashboards, etc.

camjw··on Data-Oriented Design (2018)
Best SWE book I've ever read and I don't even work on something video game adjacent any more
camjw··on Apple Reveals 'Vision Pro' Headset
It's existed for ages on other headsets and it's called Gravity Sketch (used to work there, the product is incredible)
camjw··on Ask HN: What's your favorite GPT powered tool?
Is this for real? They called it a Sentient Killer Drone?
camjw··on Brex’s Prompt Engineering Guide
This is fair, typically I supply data as compact JSON but ask for responses as pretty printed JSON which is quite a large token penalty but tends to strongly reduce malformed JSON outputs.
camjw··on Brex’s Prompt Engineering Guide
I've heard this a lot but don't understand where this idea comes from. With JSON you can strip whitespace whereas with YAML you're stuck with all these pointless whitespace tokens you can't do anything about.

I would recommend the exact opposite, JSON is just as effective while using less tokens.

This example JSON:

{"glossary":{"title":"example glossary","GlossDiv":{"title":"S","GlossList":{"GlossEntry":{"ID":"SGML","SortAs":"SGML","GlossTerm":"Standard Generalized Markup Language","Acronym":"SGML","Abbrev":"ISO 8879:1986","GlossDef":{"para":"A meta-markup language, used to create markup languages such as DocBook.","GlossSeeAlso":["GML","XML"]},"GlossSee":"markup"}}}}}

Is 112 tokens, and the corresponding YAML (which I won't paste) is 206.

What am I missing?

camjw··on The future of generative AI is niche, not generalized
Can you give a concrete example of this problem that you expect an LLM to not be able to solve? It's fine saying "abstract concept" and "problem it has not encountered before at all" but these seem to me quite fuzzy concepts.
← PreviousPage 2 of 5Next →