Pinned Tweet
.@jimydotorg and I launched HOTorNOT 20 years ago this October š® @mashable wrote a great retrospective about it. Realistically it is probably the last time anyone will ever want to write about HOTorNOT. I'm happy with this article as its final chapter. mashable.com/feature/hotornā¦
51
60
431
james hong retweeted
Possibly you suspected a couple of days could go by without a new release from us. Here we developed and then open sourced ultra fast weight updates between trainers and samplers for RL, even when trainers and samplers cross datacenters.
Large-scale RL has a problem: weight updates from trainers to samplers are slow and expensive
Weāre sharing weight teleportation: faster weight updates than existing intra-cluster techniques with internet speeds, enabling samplers anywhere (maybe even space!)
Weight icient (<0.23s sampler pause time, ~3s non-blocking compression + prepare), the bottleneck ends up being the GPU hardwareās decompression ASIC
We unlock these gains by using GPUs for massively parallel compression of model weight changes, engineering a heterogeneous decompression and patching pipeline, along with other optimizations and overlaps. These techniques are co-designed with hardware-native features in NVIDIAās Blackwell generation of chips
Weāve open-sourced weight teleportation, so everyone can use it today
5
8
77
3,973
product idea: candy dispenser for halloween that uses CV to make sure it is not dispensing to a repeat person. Can be a dumb but connected device just streaming video and have the actual software run on a laptop in the house. (saw people complaining about kids stealing entire bowls if they leave the house..)
2
0
3
583
There is some really weird panic content coming out about clouds in the last few days. I'm not sure what data these folks are seeing but I'm not seeing it at all with bigger players. The only thing I can think is that maybe tiny players are an entirely different market who can't find the demand and vice versa, and that's why there are all these brokers who also from what I hear are often full of shit on what they quote they can get. It's important to recognize that if this is the case, this is like a narrative built on a detached market with a miniscule amount of volume
The other thing I've noticed is people saying "uh oh, so the hyperscalers and SpaceX are screwed, because if anthropic had demand they would be scooping this shit up"
Bro, anthropic isn't even asking for a quote on your 20mm in hardware (that's rounding error) based in some datacenter with connectivity that doesn't fit well into their network map.
The other thing I've noticed is people saying "uh oh, so the hyperscalers and SpaceX are screwed, because if anthropic had demand they would be scooping this shit up"
Bro, anthropic isn't even asking for a quote on your 20mm in hardware (that's rounding error) based in some datacenter with connectivity that doesn't fit well into their network map.
0
0
9
911
Was just watching old viral videos that got passed around (often as flash swf files, pre YouTube). Love this classic
youtu.be/Cp-Ys_iFwnM
youtu.be/Cp-Ys_iFwnM
1
0
5
1,196
If you have a Tesla and haven't tried FSD in the last year, you need to try it again. Free month trial with this link. (i don't care about the referal credit, I'm posting this as a PSA because FSD is so awesome. )
tesla.com/get-app/referrā¦
tesla.com/get-app/referrā¦
148
130
1,237
434,469
(Apparently I hit the limit on referrals, you can only make up to 12.. so if the link above doesn't work, try one from someone else on this thread!)
1
1
4
2,309
james hong retweeted
Real artists ship. Extremely proud of the Lambda team for our absolute focus on delighting customers.
If you ask frontier lab and hyperscaler insiders, Lambda is absolutely beloved because of our track record of reliable and on-time delivery.
If you ask frontier lab and hyperscaler insiders, Lambda is absolutely beloved because of our track record of reliable and on-time delivery.
Others announce. We deliver.
Proud to be recognized by enterprises, hyperscalers & frontier labs.
7
6
66
3,581
james hong retweeted
Funding rounds aren't milestones. It shouldn't be your goal to raise a series A. It should be your goal not to need to.
476
629
7,459
381,865
Finding a better way to educate people has been one of @gharik's long term goals. An adaptive teacher that understands the student is the best kind of teacher we have, and letting that adaptation be deep and one-on-one is a huge deal. Hope people build on this!
Iāve been so excited for this release. We use our user models to train better tutors and enable models that donāt just answer homework problems, but work with a learner to study and understand a subject. Looking forward to seeing what educators, tutors, parents and others can do with this.
So much has been written about cognitive surrender and decline when people use assistants, this is a step in a different direction, one where people become better at doing things themselves even as they call upon assistance.
2
2
24
1,643
james hong retweeted
Iāve been so excited for this release. We use our user models to train better tutors and enable models that donāt just answer homework problems, but work with a learner to study and understand a subject. Looking forward to seeing what educators, tutors, parents and others can do with this.
So much has been written about cognitive surrender and decline when people use assistants, this is a step in a different direction, one where people become better at doing things themselves even as they call upon assistance.
So much has been written about cognitive surrender and decline when people use assistants, this is a step in a different direction, one where people become better at doing things themselves even as they call upon assistance.
Introducing Ampersand Tutor (&Tutor) - we trained a model to learn like people, then trained &Tutor to teach it
On a widely-used tutor model eval, &Tutor performed best, showing how better models of people can help build AI for humans
Available through our API - try it!
5
9
87
9,319
Maybe we are on a Von Neumann probe right now and this universe is just inflight entertainment.
Also, very true about Larry Page. The guy is right a lot.
Also, Bob.
Also, very true about Larry Page. The guy is right a lot.
Also, Bob.
Our civilization would be winning wondrously if we are exploring the galaxy with Von Neumann probes by 2100!
Larry Page might be right. He has correctly predicted many things.
1
0
5
835
Oh shit I thought I was smart for getting a TV antenna , I gotta switch to radio now?
@pokeroots I don't think it works that way. As I understand it, the TV signal, especially if it's via streaming, gets relayed through so many different systems before it reach the viewing screen. Radio is a simpler signal flow.
0
0
2
824
Neoclouds are REITS.
Oracle CFO Hilary Maxson says Oracle is renewing 4-year-old GPUs at 20% premiums to their original contract price
āOf course thereās a question of recontracting and the necessity for the renewal of GPUs at the end of our various contracts. Mahesh walked through the in terms of technology and how weāll continue to drive value for our customers over time. Weāre building a business for today and tomorrow, and we expect to recontract at similar margin rates or better than what we have today.ā
āIn the case of renewal of GPUs or other IT equipment, we gave a real life example in our Q1 where weāre selling or renewing 4-year GPUs at 20% premiums to their original contract price. We put that information into the slide here. We think thatās a powerful example of the potential upside from the GPUs we have in our portfolio today. We wonāt count on it, but we do think that these are valuable assets both for today and into the future.ā
0
1
8
907
james hong retweeted
Thrilled to announce the general availability of
@positron_ai's Atlas Inference Appliance on
@OracleCloud, making Positron the first AI chip startup to be deployed at scale with a hyperscaler. Look out for more announcements this week at Oracle Cloud World! blogs.oracle.com/cloud-infrastrā¦
@positron_ai's Atlas Inference Appliance on
@OracleCloud, making Positron the first AI chip startup to be deployed at scale with a hyperscaler. Look out for more announcements this week at Oracle Cloud World! blogs.oracle.com/cloud-infrastrā¦
14
15
144
12,700