@ntkeep

Previous: Co-founder https://xitter.cf/t.co/WtGHhZyHvY, Head of audio software @ Meta, co-founder of @twobig_ears (acquired by Meta) + more.

 UK (sometimes the Bay Area)
re-sounding.com
Joined November 2009
  • Tweets14.8K
  • Following193
  • Followers2.2K
348Media
Varun Nair retweeted
Weavable AI@weavable_aiMay 11
Hello World! We are on Product Hunt. AI agents are only as good as the work context they're given. Right now there's no layer that tracks cause and effect across your work data over time. Weavable is that layer. One MCP endpoint, continuously updated. producthunt.com/products/wea…

Weavable: Give every AI agent persistent work context | Product Hunt

Weavable gives AI agents persistent, live work context from the tools your business already runs on. Through a single MCP endpoint, it turns scattered updates, relationships, and system changes into...

producthunt.com
1
1
1
54
Just launched Weavable on Product Hunt. AI agents are only as good as the work context they're given. Right now there's no layer that tracks cause and effect across your work data over time. Weavable is that layer. One MCP endpoint, continuously updated. producthunt.com/products/wea…

Weavable: Give every AI agent persistent work context | Product Hunt

Weavable gives AI agents persistent, live work context from the tools your business already runs on. Through a single MCP endpoint, it turns scattered updates, relationships, and system changes into...

producthunt.com
4
1
6
142
When using foundation models for product features, the bottleneck shifted from "how do I train a model?" to "how do I prepare data that actually works?"
1
87
Better data architecture beats bigger context windows every time.
55
LLMs care about how you structure information, not just what you feed them. What you exclude matters more than what you include.
90
A startup really is about all your good decisions making the bad decisions look worse
154
I wrote about how I’m using our own product to move faster and to quickly distill what I need to focus on.
1
1
140
Link here linkedin.com/feed/update/urn…

Distilled Decision Making Every morning at 9am, post-workout, I receive an email from thatworks.ai...

Distilled Decision Making Every morning at 9am, post-workout, I receive an email from thatworks.ai distilling thousands of data points into what actually needs my attention: key HubSpot activity,...

linkedin.com
71
Context is a bottleneck when working with LLMs, not prompt engineering.
1
1
124
More here linkedin.com/posts/varun-nai…

#llms #ai #product #startups | Varun Nair

Context is a bottleneck Most discussions about LLMs focus on their capabilities. But capability isn't a bottleneck–context is. "Context" is the data you provide to an LLM to work with: the informat...

linkedin.com
48
Startups are the best way to make a real impact with research. But how do you get started? We’re hosting events at Edinburgh, Cambridge, Oxford, Manchester, Newcastle, and Warwick to dive into the scientist-to-founder journey and how deep tech companies get started. 👇
5
17
4
70
14,873
Can I downrank posts with the word “wild”?
1
134
More here: no10innovationfellows.campai…

No10 Innovation Fellowship

Deploying world class technical talent at the heart of government. Hosted by the No10 Data Science team, fellows bring knowledge, skills, and experience from across industry and academia for high-i...

fellows.ai.gov.uk
2
3
35
5,628
Lots of energy recently in UK tech re making government better... so here's how to help: apply to be a No 10 Innovation Fellow and do a tour of duty working on some of the country's most complex challenges. Huge opportunity for impact
18
66
16
309
66,551
My latest weekly post: garbage in, garbage out. Dumping data into a LLM's context is like recording in a noisy room – you end up with competing signals that steer the output into vague generalities. So, how do you maximise the signal to noise ratio?
1
2
105
Originally posted on LI: linkedin.com/posts/varun-nai…

Garbage in, garbage out. Working with LLMs reminds me of my past life in audio engineering. Want...

Garbage in, garbage out. Working with LLMs reminds me of my past life in audio engineering. Want great sounding audio? Start with a great microphone! The same principle applies to LLMs. Just dumping...

linkedin.com
47
I’ve been pretty happy with all the tech choices we’ve made at thatworks.ai. We use a standard react/nodejs/k8s stack but some of these older JS libs are just terrible. For nodejs, we replaced webpack with esbuild and our build time improved by 120x! create-react-app 🤢 is next.
81
1500kg per *hour* is incredible!
Boyan Slat
@BoyanSlat27 Oct 2024
So far in 2024, @TheOceanCleanup has been collecting an average of 1500 kg of trash for every hour of the year.
2
306
A significant amount of prompt engineering effort is just trying to get the LLM to “sound” how you’d like it to without resorting to fine tuning or a large corpus of multi-shot examples.
Andrej Karpathy
@karpathy6 Oct 2024
Not fully sure why all the LLMs sound about the same - over-using lists, delving into “multifaceted” issues, over-offering to assist further, about same length responses, etc. Not something I had predicted at first because of many independent companies doing the finetuning.
166
TIL that the House of Commons library is an excellent resource for a range of topics, like this one on budget deficit: commonslibrary.parliament.uk…
76
Load more