hustwindmaple1

Every major Cloud vendor is trying to develop their custom AI ASIC. Putting Google aside, Amazon has trainium/inferentia, which Anthropic uses quite extensively. Microsoft is doing sth. similar, although they are quite behind. OpenAI is doing it. Meta is doing it. That's why the stock price of Broadcom/Marvell soared.

I remember Kumo was focusing on GNN when it was founded (Jure's strength back then). Looks like they are pivoting or have pivoted.

You are basically doing a heurstic. Your solutions are not guaranteed to be optimal. Integer programming is the way to do.

If you are not a paying GCP user, there is really no point to even look at Vertex AI.

Just stick with AI Studio and the free developer AI along with it; you will be much much happier.

Really appreciate your team's enormous efforts in this direction, not only the cutting edge research (which I don't see OAI/DeepMind publishing any paper on) but aslo making the content more digestible for non-research audience. Please keep up the great work!

His Alibaba/Arm bets are legendary. Both made $100B+

Housing is definitely not easily affordable. But the other points are valid.

Cool blog

there is limited TPU support in pytorch via torch_xla

When he drops a vid, you don't ask questions. You watch first and then ask questions :)

High-Flyer is pretty damn rich actually. Someone did some calculation and it turns out they are spending ~$200m (somewhere in that range) on 150 employee compensation alone.

prob. not prof; just phd students needs pubs to graduate

Are you comfortable to send your personal/company code data to a Chinese company (ByteDance)?

There are a couple, i.e., OLMo 2

Exactly. Kind of like the old days when they put together a massive amount of commodity CPUs to build search.

GDM is a research lab. They are not set up for production. There are other teams in Alphabet doing productionization stuff.

Large Chinese companies usually have overseas subsidiaries, which can buy H100 GPUs from NVidia

well, is Waymo doing better than the PyTorch-powered Tesla?

Just use Brave. You'll thank me later ...

Reformulation is a good strategy for many hard problems.

It might also be possible to add problem-specific cuts/heurstics to the solver so that it can solve it fast.

I was on CPLEX team for a few years. Core developers are all PhDs from Stanford/MIT and etc. So it's very hardcore stuff, no less than AI research.

Completely agree that size is not a good proxy for estimating MIP difficulties. Internally we colllected a bunch of very hard problems to sovle from different domains. Some are actually pretty small, say a few thousand variables/constraints. IMHO what made hard problems difficult to solve is actually the 'intneral structure' of the problems. And modern industry solvers all have a lot of built-in heurstics to take advantages of the structures, i.e., what kind of cuts, presolve/diving/branching strategy to apply, how to get the bounds ASAP.

Interestinglly at some point some folks even tried using machine learning to predict strategies. Didn't work quite well back then (10+ years ago). There was some work of using seq2seq for MIP (pointer network, I think) a few years ago; worked OK. So I'm really looking forward to some breakthroughs by LLM.

It's a shame that after IBM aquired ILOG (which owns CPLEX), most of the ppl left for Gurobi.

The problem is that there is no bette alternative.

Is Clubhouse still alive?

Right, but they can just use Llama/Mistral for free, instead of their inferior models, which I'm sure take quite a bit of resources to train in the first place.

I wonder why companies like IBM are jumping on the LLM bandwagon and training/releasing models that have no chance of competing with Llama/Mistral? To me it just looks like a complete waste of $$ because nobody will use them in any serious scenarios

It's MQA, documented in the tech report

maybe they are tracking # of accounts for their next raise from VCs

It's not their emails. It's all the ads you see on YouTube, display ads on random websites through Google/FB ads network, ads within xyz apps, and etc.. They spent a lot of $$$ on those ad networks.

Those are usually targeted through personal email addresses and device IDs. Impossible to opt out on users' end.

I emailed their customer support and asked them to stop targeting my email account in all their ads. They said they did, but I still got a ton of their ads afterwards. Anybody with ads experience would instantly know their ads targeting wasn't working great and they were wasting a sh*t ton of money.

One of the most annoying companies that ad spams like crazy in 2020/2021. i literally had to beg them to stop sending ads my way.

Now they are running out of money and have to cut staff. How ironic!