m_ke

14 ms·

151.
by m_ke 2y ago
Those models will be here within a year. Long context is practically a solved problem and there's a ton of work now on test time reasoning motivated by o1 showing that it's not that hard to RL a model into superhuman performance a
152.
by m_ke 2y ago
I've been thinking about this a bunch and here's what I think will happen as cost of writing software approaches 0: 1. There will be way more software 2. Most people / companies will be able to opt out of predatory VC funded
153.
by m_ke 2y ago
Ray is a distributed computing framework that has a fast scheduler, 0 copy / serialization between tasks (shared memory) and stateful "actors", making it great for RL but it's more general than that. I'd recommend c
154.
by m_ke 2y ago
Yeah I'd never go for spark again, all of its use cases are better handled with either DuckDB or Ray (or combination of both).
155.
by m_ke 2y ago
There's a ton of recent work on data curation / synthetic data generation that shows that smaller high quality datasets go a lot further than scaling up on noisy web data. The scaling law plots are log scale so to get more juice w
156.
by m_ke 2y ago
Was it access to cognition_labs repos or a github key that gave access to customer repos?
157.
by m_ke 2y ago
What actually happened?
158.
by m_ke 2y ago
No it's not, anyone who ever used docker knows that the images don't get committed and setting random seeds is ML 101 for anyone who is trying to run repeatable experiments.
159.
by m_ke 2y ago
He's a graduate student with multiple influential papers, including NeurIPS best paper award here that might change how to do image / video generation, I think he can figure out docker and tweak a few lines of python.
160.
by m_ke 2y ago
Who is "we" in this statement?
161.
by m_ke 2y ago
I used to work at one of the first deep learning image recognition startups, it was all going great, we raised a ton of money, hired a lot of people and built some amazing technology. The founder was on the covers of fancy business magazine
162.
by m_ke 2y ago
hey just got back on and the papers you shared are the main works that I was about to link. There's also a new VLA paper fro waymo https://arxiv.org/abs/2410.23262v2 and some recent talks on youtube: - OpenVLA: h
163.
by m_ke 2y ago
I did a review of state of the art in robotics recently in prep for some job interviews and the stack is the same as all other ML problems these days, take a large pretrained multi modal model and do supervised fine tuning of it on your dom
164.
by m_ke 2y ago
I've been using 42" 4K TVs as my monitor for like 10 years now. 2 years ago I upgraded to an OLED LG A1 and it has been amazing. https://www.rtings.com/tv/reviews/lg/a1-oled For anyone using a TV I
165.
by m_ke 2y ago
https://developers.facebook.com/blog/post/2021/09/07/eli5-op... That's exactly what they id with their server design. I'm saying come up with an open standard for tensor processing chips,
166.
by m_ke 2y ago
Half of NVIDIA's 2nd quarter revenue (30 billion) came from 4 customers, with Microsoft and Meta already having spent 40-60 billion each on GPU data centers (of which most goes to NVIDIA). "Open"AI just raised a few billion a
167.
by m_ke 2y ago
I wonder when Meta, Microsoft and OpenAI will partner on an open chip design to compete with NVIDIA. They’re all blowing billions of dollars on NVIDIA hardware with like 70% margin and with triton backing PyTorch it shouldn’t be that hard t
168.
by m_ke 2y ago
I wasted two days a few months ago when Amazon randomly started charging us $1K a day for static s3 buckets with a few GB of data that hasn't been touched in like 4 months. Turned out it was a billing error on their side that they woul
169.
by m_ke 2y ago
With these fancy chips, can they please let me use it as my primary computer so I can ditch my macbook?
170.
by m_ke 2y ago
Imagine a indie hackers founders cooperative, you take a few successful bootstrapped founders and start a new holding company that shares an office space and has staff to handle legal, accounting and all the other stuff that most engineers
171.
by m_ke 2y ago
Yeah I'd imagine it would have to be a single holding company that has members join, instead of investing for equity. So as an example you could have hypothetical opensource.coop that funds open source projects like signoz or posthog,
172.
by m_ke 2y ago
It's a shame there's not a YC like incubator for tech worker coops, cut out the VCs and have milestone based funding in exchange for a percentage of future profits that would be reinvested in future projects.
173.
by m_ke 2y ago
I wonder what percentage of that is people getting on a bus to get to a train, where they actually pay the toll. I used to have to take a bus to a train and it was normal to have people get in through the exit doors in the back because the
174.
by m_ke 2y ago
Similar story about hotel price fixing: https://popular.info/p/how-hotels-are-colluding-to-jack
175.
by m_ke 2y ago
Can't wait to hear how he's a communist for providing kids with free lunch for the next 8 weeks.
176.
by m_ke 2y ago
Because he doesn't want to deal with sama
177.
by m_ke 2y ago
It would be a lot harder to do if the market had more viable competitors. In a world with 5-6 dominant options and a long tail of niche search engines you'd probably have options that penalize blog spam, aggregators like pinterest and
178.
by m_ke 2y ago
It gives people room to try, especially now with new entrants like perplexity that might offer better results. Google search went to shit and ruined most of the web with it, by creating an SEO industry that's 100% focused on optimizing
179.
by m_ke 2y ago
And the funny part is, these LLMs are amazing at writing YAML config files. I always just let it write my first draft of docker, k8s and terraform configs.
180.
by m_ke 2y ago
In the CNN boom days after alexnet the playbook was: take your research lab, slap a c crop and new logo on it, keep doing your research and get acquired by FANG for high 8 figures. Only difference in this cycle is that you can't do rea

More ›