Thank you to everyone who participated in the Cloud Run Hackathon with Product Hunt, and congratulations to our winner, Inferock! 🎉
Learn more about Inferock → https://t.co/YtecmbOmwt
Hey guys, thank you for all your support we won the @GoogleCloudTech hackathon yesterday.
To celebrate we're hosting an event in San Francisco on October 22nd 6-8 PM. Join us if you're in town.
Please register here: https://t.co/JRQfN7zldz
I look forward to meeting you!
cc: @RajSetty@bharathakoneti@iamgangesh@himashwetha12
We launched Inferock today, and we're currently #2 Product of the Day on Product Hunt!
Thanks to everyone who's checked it out and sent feedback!
Failed model calls can still cost you. Inferock shows what happened and what you were charged. And for eligible, verified failures that were billed and not recovered, Inferock Watch accounts can earn credits back to use in the future on Inferock Serve.
So you get more than a failure report. You get a record of the call and, when it qualifies, credits toward the next one.
We're live at: https://t.co/fNpjRIPsdB
Check us out on ProductHunt: https://t.co/i45ojzTepT
You picked a state of the art model. Your users got a spinner, half an answer, and a retry.
Good thing the benchmark score was excellent.
We look at Inferock’s 19 quality requirements for the part that actually matters: the service delivering the model. https://t.co/bMql7S8dlM
JavaScript's MOAT is the incredible ecosystem built around it.
Most teams have switched to giving agents most of the work, they can work in either JavaScript or any other specialized language all the same.
The node package ecosystem that JS built over the last decade and a half is not replicable.
Training from scratch requires a massive amount of information. In most cases, it's better to post train with domain specific data.
The common user is better off using off-the-shelf APIs that provide transparent and reliable metrics.
That is something we're solving at https://t.co/gTAO7Q8uqx
@ChanduThota This is a great change.
Google Docs is the most accessible and easy to use text editor.
It being able to work with direct markdown solves many conversion issues of old.
23x more traces per dollar is insane!
I’d love to see how that holds up across different production workloads.
We’re also building something similar, https://t.co/gTAO7Q8uqx, to help teams keep an eye on model performance, usage, and spend across their providers and maintain transparency and accountability across providers.
@mdnlabs We recently took help of AI to build the basic outline of our discord server and had it work on setting up the roles too.
There is a concern for abuse but for small tasks, it is incredibly useful.
@TheNathanRuff AI/ML has always been around, LLMs just made them accessible to the general public.
Much of core ML theory was written in the 20th century.
@ClementDelangue Agreed.
That’s a big part of why we’re building Inferock Watch, to make model performance, usage, and costs visible in your own provider accounts.
AI needs more transparency where people actually use it.
We wrote about why an LLM stream starting doesn’t mean it finished, and what your app should check before calling it a success. https://t.co/7im8tPw4Ys
@getrelayen@sama Hi Saq, saw your comment.
I’m with Inferock. We measure response checks, timing, usage and available cost data per call, and our dedicated option is priced under agreed terms before provisioning.
If you’re building with models, might be worth a look: https://t.co/q17L5LXhsf”
@jon_stokes Hi Jon, saw your thread.
I’m with Inferock. We’re building per-call measurement for response checks, timing, usage and available cost data, plus dedicated capacity with price and service terms agreed upfront.
Thought it might be relevant: https://t.co/q17L5LXhsf