Pinned Tweet
Excited to share @agency - a better way for companies to care for their customers
We've raised $12M from @Sequoia, and HubSpot founder @bhalligan is joining our board
Hear more from my co-founder @eliast below covered by @MTemkin @TechCrunch
Luke Van Seters retweeted
Today, we’re announcing Antimetal’s $20M Series A.
Writing code is no longer the hard part. Maintaining it is.
We’re automating everything that happens after you deploy.
We have 60 spots available. Reply for early access.
Luke Van Seters retweeted
my friend @eliast is in his 40's, looks like he's in his 20's, and has the energy of a teenager... and we @sequoia are delighted to be in business with him for the third time!
@MTemkin does a nice job of capturing what he's up to here:
techcrunch.com/2024/10/16/af…
Luke Van Seters retweeted
Today’s the day. Introducing my new company, @Agency — backed by @Sequoia and Hubspot Ventures; with my friend and mentor @BHalligan on the Board. Enjoyed getting to talk with @MTemkin @TechCrunch about what we’re building.
agency.inc/
We're hiring a small, exceptional team of mostly engineers.
Send me a DM to learn more
techcrunch.com/2024/10/16/af…
Fun fact for keyboard maxis
You can push your key repeat faster than the system settings limit on OSX
defaults write -g InitialKeyRepeat -int 10
defaults write -g KeyRepeat -int 1
(Requires a restart)
Luke Van Seters retweeted
A slow Postgres query on launch day can be a nightmare scenario that jeopardizes your AI application. Learn @lvanseters effectively managed a launch day outage for zelma.ai when a huge spike in traffic, coupled with unexpected queries caused Postgres to peg to 100% CPU.
Shout out to @htormey for being a great host!
Lots of fun things we talked about:
- Getting known-good SQL from LLMs
- Speeding up LLM calls (got a 10X speedup on zelma)
- Vector DB tradeoffs
- Self-hosting
- Constrained grammar generations
- A DB war story
- and more!
Thinking about building an AI powered app to explore a large data set? Want to know the pro's & con's of using pgvector over a dedicated vector store? Interested in hearing tips for reducing LLM latency? If so checkout my interview with @lvanseters about how he built Zelma.
stepchange-blog.ghost.io/how…
Idk about the feature, but I like the idea of mimicking a call
You always see people demo voice features yelling into their phone and having it yell back - which makes no sense in public
Taking it like a call feels like an obvious answer hiding in plain sight
It seems GPT-3.5 is at least as good or better than GPT-4 for many "agentic" tasks
Left is a talk from Andrew Ng - shows looped LLM approaches on HumanEval (coding) with 3.5 barely edging out 4
Right is from Gorilla - an LLM tool use paper (up to the left is better)
Sources:
What's next for AI agentic workflows ft. Andrew Ng of AI Fund youtube.com/watch?v=sal78ACt…
Gorilla: Large Language Model Connected with Massive APIs arxiv.org/abs/2305.15334
OpenAI shared a customer story for our work with @ProfEmilyOster to build Zelma.
Her team painstakingly gathered and cleaned all historical US grade school test results.
So to make the data accessible, we built an LLM-powered app where users can explore the data just by asking.
