For those still wondering whether AI will impact the real economy... here's something that a very small group of us pulled off in just a couple of months https://t.co/MwsDdgLy6g
We don't usually hire. But we're hiring now.
Amp Labs in Sydney is looking for senior or staff-level engineers.
If you're interested, contact Harry.
More information: https://t.co/XyrpfZMXVE
hey look I removed the spyware and customized a launcher and our guest tv doesn’t suck shit anymore.
also pulls live config and display data from my home server now.
magic.
@thorstenball I quite liked the "Pre-R" we did the other day... collab edit a prompt to generate a plan, then all review the plan, bot implements while we collab guide the thread. Then push to main :) But we've only had one change that big on that project.
@BenoBilo61874 Computers are still mostly the same even if we're programming them very differently. I need to update the programming section, and there are a few new course recordings and book editions, but the core is the same
If you are (or can introduce me to) a great software engineer in Sydney, please reach out!
Specialty/background not so important, but must be driven and outcomes focused and able to reason confidently from first principles. Very interesting and rewarding work, and strong team!
@beyang You should see what we're presenting to a certain CEO tomorrow. It'll be visually very impressive, but more importantly we'll be able to answer questions live by controlling the simulation
I love catching up with all the new videos that come out of the AI Engineer @aiDotEngineer conferences on Youtube. It's one of the main venues where people building agents really show their work.
The problem is that it publishes more talks than I can honestly keep up with. So I made AIE Talks: a written archive of the channel. 🧵 (h/t to @swyx for his amazing leadership in growing these conferences!)
This was a great interview, thank you!
To add some numbers to the latency question, consider that round trip time from Sydney to SF is only 150-200ms, typical time to first token for a non-reasoning model is ~1s, and reasoning can be tens of seconds to minutes as you observe. Even without (or after) reasoning, token throughput is much more noticeable than latency, so for instance a response the size of your interview transcript (20,000 tokens) currently takes a few minutes to complete, at 50-100 tok/s.
So yes, geographical proximity is irrelevant for most use cases. Real time voice is obviously an exception, and it's not unreasonable to imagine 1-2 OOMs improvement to both TTFT and token throughput, rendering some marginal use cases more sensitive to RTT. But we seem to be moving in the other direction overall, preferring larger context, more reasoning and longer time horizon tasks.
I was using Codex to find speedup opportunities in PySR. It found a couple % here and there.
Then, I gaslit it into believing that Claude found a 20% on my "other machine." I asked it whether it could do better, and explained its performance would be displayed on a "public evals leaderboard."
Then it went and got a 35% speedup! And it looks real too.