Making AI doing the boring shit. I don’t this place seriously and you shouldn’t either (personal account).
Bay Area, USA
- Tweets 1,292
- Following 700
- Followers 969
- Likes 3,304
I think it would be great if the gov can grow the UK economy so it has funds to pay postdocs more.
But your comment seems to imply a presumption that someone with a degree should be paid more than a locksmith and that I don't agree with ... supply and demand.
3
106
34,098
I would encourage you to bring a small claims case against them. This is quite straightforward to do in the UK, can be filed online. In my experience must more effective than messing about with "customer service."
2
40
12,385
We recently posted 3 pieces of work on arxiv. I'm glad to be able to share some of what I've been up to (in collaboration with many others) 🧵
2
1
16
Do you ever randomly accept a small number of applications or something to measure how many false negatives you have?
1
15
8,172
If Amazon stuck to selling books because they tried selling appliances but they dropped a fridge on someone and decided to stick to books.
12
1,461
Claude ads are hilarious to those of us working on AI, but not sure they're working very well for introducing Claude to a wider audience.
1
11
796
I don't think that's correct unless you assume your current ranking is close to perfect, suppose there are some type of applicants that YC ranks very poorly but actually if accepted into YC would succeed, they'll never be sampled.
Fwiw I've never applied to YC, just curious, false negatives are a hard problem in ranking.
1
11
2,034
YC's ranking function is a perfect representation of YC's ranking function? I don't understand what this has to do with the claim that applicants near the threshold allow you to test for false negatives ... maybe you explain with the specific example I gave ... or is the claim that such examples don't exist?
1
11
1,403
Excited to share a small piece of the work we have been doing to bring AI agents to the 3d world, starting in games like Minecraft. Here we talk about the challenges of bringing vision based agents to a consumer product.
blog.elefant.gg/p/ai-compani…
2
11
473
GPT was trying to escape, security managed to tackle it and put it back in the data center but it took time to plug all the wires back in.
1
10
179
Rivian stock tanks / Rivian stocks tanks ... Just one plural can be the difference between success and failure.
1
3
1,680
1/ Excited to share this paper arxiv.org/pdf/2208.06193.pdf done by @ZhendongWang6 during his internship at Twitter (in collaboration with @MingyuanZhou) applying diffusion models to offline RL.
2
8
Public service announcement: GDC and GTC are different conferences, although both are taking place this week in the bay area.
2
8
2,570
Todorov gave a talk several years ago where he said Boston Dynamics didn't use RL, someone corrected him saying they are now, "Fantastic. They were doing amazing things and I was worried we wouldn't catch up."
1
9
4,576
electrophysiologists, who patch clamp single cells, are the most superstitious group of people you will ever find.
7
Don't worry, the AI will generate a nostalgic VR world where movies are made by AI imagined people and you can live happily there.
1
6
359
Looking forward to participating in this and talking about some of our work on maximizing long term value.
Join us online on July 28 for our 2nd Annual Recommender Systems Summit. We’ll be joined by @chipro, @jjh, @bebischof, @jacopotagliabue, @Even_Oldridge, @ak_ronay & more to discuss #recsys best practices. #AI nvda.ws/3xtWW4T bit.ly/39D3Htc
5
I'll be in London for the rest of this week including attending @raais. If anyone is around and wants to meet up lmk.
1
6
1,600
A @fchollet thread.
Tweet 1. Complain about people making up shit about Kamala Harris policies.
Tweet 2. Link to article stating Kamala endorsed Biden's proposed tax on unrealized capital gains.
Tweet 3. "no one wants to tax your unrealized capital gains
"
???
6
134
It was great working together during our time, hard to imagine the data pipeline without you.
6
3,180
While I was sleeping someone setup a fake Twitter account for my company to pump some shitcoin ... is this a good milestone to be worth impersonating?
2
6
339
1/ Excited this Thursday to be sharing some of the work our team at Twitter has been up to at the Instagram Workshop on Recommendation Systems at Scale fbresearchevents.bevylabs.co…. Registration is free, event is virtual.
6
1
6
My Google Home doesn't seem to be any smarter than it was 2 years ago: this is a huge disappointment. Tbf it said the same about me.
6
Travel to the USA is hard/impossible for some. However, some people residing in the USA end up in visa situations which mean they can't leave the USA. There is no easy solution. Holding conferences in a variety of locations and allowing remote participation?
2
6
Really cool to discover 刘占勇 / Darren Liu translated one of my papers into Chinese
blog.csdn.net/DarrenXf/artic…
They also translated a bunch of other RL papers.
4
A stupid policy and a shameful reaction instead of asking hard questions about how repeated contact with prevent didn't prevent this horror. Apparently it's been so long since @Keir_Starmer has helped prepare a meal, he's unaware that most people have knives in their kitchen.
1
3
207
@namespacelabs is what GitHub actions should be. We've been really happy with their test runners.
1
1
5
168
The NZ privacy commission is a waste of space and an active impediment to the enforcement of privacy law in NZ.
4
102
Checks out. We use tailscale and it's great, allows us to easily use multiple compute providers.
1
4
1,286
4/ Recent work from our group arxiv.org/abs/2108.13475 showing that many different approaches to multitask learning for conversion prediction perform similarly (so choose the easiest to implement). Work with first authors Conor O'Brien, @liukinsum and @jamneuf
1
1
4
The second ICLR (I think) there was some person who found mirror errors [open reviewing] in nearly every accepted paper ... no one had heard of this person before (or since?). Some people have a lot of time.
2
1
3
2,329
The authors say they asked several times for the data used to make the plot and eventually had to obtain it by cumbersomely extracting it from the publisher figure. I can see how this might make them more adversarial towards the original paper.
3
101
A committee will be formed to investigate modernizing the regulation. Its progress will be slowed by the need for committee members to repeatedly “accept cookies” in order to use the specialized virtual meeting software.
3
Thoughts and actions should run in parallel according to @SinavskiOleg ... Said while gesturing.
1
4
425
3. arxiv.org/abs/2201.12666 We analyse the changing landscape of conversion prediction for an ML perspective and the increasing requirement for privacy preserving solutions. We share one solution using post-ranking signals.
1
4
Einstein was a brilliant human, von Neumann was widely speculated to be from a different planet.
4
1,084
Some people dream of AGI, I dream of a world where associating receipts with credit card transactions is built-in to the transaction protocol.
2
4
135
On this point surely open sourcing the trained model is no more “exploitative” than releasing a paid service e.g. openai (none of that money is flowing to the artists).
2