"There are far fewer East Asian CEOs in the Fortune 500, and most of them are the founders of their companies like Jensen Huang (Nvidia), Tony Xu (DoorDash), Lisa Su (AMD)"
I think the author has been using an LLM to extend the text or get facts and it's making weird phrases and/or mixing statements together. It's specially visible on the 'The Costs of Conflict Avoidance' section when the theme changes for no reason and it starts listing "things to do to avoid conflict".
In the case you're mentioning, in the Indian CEO list he lists FORMER Indian CEOs:
> Among the Fortune 500, the CEOs of Alphabet/Google (Sundar Pichai), Microsoft, (Satya Nadella), Adobe (Shantanu Narayen), Chanel (Leena Nair), IBM (Arvind Krishna), Micron (Sanjay Mehrotra), Palo Alto Networks (Nikesh Arora) and former CEOs of Mastercard (Ajay Banga) and Pepsi (Indra Nooyi) were all born in India and were appointed the CEO position.
In the East Asia CEOs list, he mixes together the "founders" list with others, misses a bunch of East Asia CEOs and creates a weird phrasing:
> There are far fewer East Asian CEOs in the Fortune 500, and most of them are the founders of their companies like Jensen Huang (Nvidia), Tony Xu (DoorDash), Lisa Su (AMD) and Hock Tan (Broadcom). Three of these are in the semiconductor industry and two of them founded their companies. This is just a list of the Fortune 500 CEOs [...]
I think they could release non-agentic models that are as good as 4o, and have almost no repercussions on sales tbh.
I have Ollama installed (only a small proportion of their clients would have a large enough GPU for this) and have download deepseek and played with it, but I still pay for an OpenAI subscription because I want the speed of a hosted model, and never mind the luxuries of things like Codex's diffs/pull request support, agents on new models, deep research etc. - I use them all at least weekly.
I pay for Cursor, OpenAI and kimi (to use with Claude Code), OpenAI is good with quickly refining my thoughts, Cursor’s subscription I’m reconsidering to cancel bought it for Claude but the rate limits are making it impossible for me to find it useful. Kimi is what truly surprises me, Claude code shows this conversation costed you $500 (based on Opus usage which is mapped to kimi k2) while I’ve barely spent $2. I have Ollama as well, majorly to quickly test small models that could be improved for our usecase through finetuning.
I've been using Kimi with Roo via OpenRouter and have been very surprised at how capable it is. It's the first open model I've tried that actually lives up to claims I see online that's it on par with this or that previous gen proprietary model. Context window has been the only negative, at least with the providers OpenRouter has been giving me but forgiveable given how absurdly cheap it is.
Considering my personal heavy use also not leading to rate limits and what I've seen by some users over the past months, I suspect a mix of actually thinking about your code before writing a prompt, managing context by documenting and running stuff like git, npm install, etc. yourself instead of "Hey Claude, setup React with Radix and install a few packages". I have genuinely seen someone use ultrathink for setting up a starter repo hosted on Github, despite the commands being listed in the readme, so I can see how certain people may hit the limits quicker than others. Still, I will cancel my Claude Max subscription if they remain intransparent concerning the amount of use we actually get, especially regarding the mail they sent out recently which stated that 20x Max users do only get 10x in terms of expected usable hours. Same goes for still not providing an official way to track how much use one has left in a week.
> running stuff like git, npm install, etc. yourself
Ah; this definitely makes sense! I do this myself and then paste back only the relevant part of the log so as to limit this. I suspect I am being more conservative than others.
They would definitely have sales repercussions, but it might be worth it.
They are fully trying to be a consumer product, developer services be damned. But they can’t just get rid of the API because it’s a good incremental source of revenue, and thanks to the Microsoft deal, all that revenue would end up in Azure. Maintaining their API is basically just a way to get a slice of that revenue.
But if they open sourced everything, it might sour the relationship more with Microsoft, who would lose azure revenue and might be willing to part ways. It would also ensure that they compete on consumer product quality not (directly) model quality. At this point, they could basically put any decent model in their app and maintain the user base, they don’t actually need to develop their own.
At the bottom it references a GitHub where people have previously added signatures against Jon Pretty - and now the maintainer says "NOTE: This repo is closed. Do not open issues; they will be summarily closed and ignored." - i.e. telling people they shouldn't even TRY to amend their signatures.
Regardless of what you think of Jon Pretty, how is this justifiable? Telling people they can't unsupport something because you're not open to issues, but also not removing it?!
He's not "obligated" to do anything but it's still immoral to abandon maintenence of something like that. If he can't be bothered to maintain it, then he should delete it.
I don’t know if the allegations against Jon Pretty were valid or not, but those who piled on against him can’t escape accountability for mob behavior (assuming Pretty was innocent) if it becomes embarrassing. At most they can say “I supported this but no longer do”, not expunge all traces.
> If he can't be bothered to maintain it, then he should delete it.
Not necessarily, plenty of projects have been put in an archive state because they are 'finished', superseded, forked, etc. This isn't code nor a living document, it was a one-off operation.
> He's not "obligated" to do anything but it's still immoral to abandon maintenence of something like that. If he can't be bothered to maintain it, then he should delete it.
Morality is subjective (that's why we have courts; which don't respect the individual and differing moralities of the parties involved, it has its own moral bar, for better or worse).
In this case, I feel it is more moral to record all the members of the mob. Maybe this would cause them to think twice before joining the next mob.
I mean, if we are going to have witch-hunt mobs, then the lesser evil is to not allow anonymous mobbers.
It's interesting looking at the messages of recent commits of people removing their names:
- Upon reflection, I don't think this letter was the right approach for this situation. Although I cannot retract my initial decision to sign it, I would appreciate having my signature removed from the document.
- We had good intentions and reasons for concern, but there was no due process, and the consequences of that can be awful. Please accept my withdrawal.
- The goal of providing safe spaces is laudable and necessary, but I expected to see further process outcomes from this effort. Perhaps some sort of SIP or scalarum iustitiae processus.
- I no longer believe the way this letter was the right way of dealing with the situation. And while I cannot undo signing it, I would like to request removing my signature.
Yeah but because it's a GitHub repo is has an inherent audit trail for that, so it's not really erasing misdeeds... indeed it highlights those people in diffs!
FWIW that statement (“do not open issues”) was added over one year ago, but the owner has also approved pull requests removing names as recently as 8 months ago.
So I think pull requests are still accepted, but issues are not.
No but 64 GB of unified memory provides almost as much GPU RAM capacity as two RTX 5090s (only less due to the unified nature) - top of the range GPUs - so it's a truly exceptional laptop in this regard.
I understand, but that is not exceptional for a Mac laptop. You could say all Apple Silicon Macs are exceptional, and I guess I agree in the context of the broader PC community. But I would not point at an individual MacBook Pro with 64 GB of RAM and say "whoa, that's exceptional." It's literally just a standard option when you buy the computer. It does bump the price pretty high, but the point of the MBP is to cater to higher-end workflows.
To emphasize this point further, at least with my efforts, it is not even possible to buy a 64GB M4 Pro right now. 32GB, 64GB, and 128GB are all sold out.
We can say that 64GB addressable by a GPU is not exceptional when compared to 128GB and it still costs less than a month's pay for a FAANG engineer, but the fact that they aren't actually purchasable right now shows that it's not as easy as driving to Best Buy and grabbing one off the shelf.
They're not sold out—Apple's configurator (and chip naming) is just confusing. The MacBook Pro with M4 Pro is only available in 24 or 48 GB configurations. To get 64 or 128 GB, you need to upgrade to the M4 Max.
If you're looking for the cheapest way into 64 of unified memory, the Mac mini is available with an M4 Pro and 64GB at $1999.
So, truly, not "exceptional" unless you consider the price to be exorbitant (it's not, as evidenced by the long useful life of an M-series Mac).
thank you for providing that extra info! i agree that $2000-4000 is not an absolutely earth shattering price, but i still wonder what the benefit one receives is when they say "2.5 year old laptop" instead of "64GB M2 laptop"
CapCut is one of several low-barrier video editing apps generally geared towards content creators/ people targeting Instagram reels, TikTok, and short video platforms.
I don't think you know what "cringe" means... this really makes you cringe? And they didn't say it was crazy, they said it was interesting enough to raise eyebrows. Everyone knows there are amazing coders in hedge funds, but not many hedge funds have forked a language - it is worthy of discussion here.
I think you're reaching for an interpretation of what I said that's easier to challenge.
Once you explain to the average person what a compiler is, they can draw a straight line from writing the compiler to the hedge fund reducing risk, gaining a competitive edge, whatever.
But this writer is choosing to hold up custom compilers as even more of an oddball move than counting cars, or shipping gold overseas. It's lazy writing from someone who is mistaking their lack of technical knowledge for common-sense insight into how strange software is. It is cringe.
I don’t understand how this works with systems like toothpaste? I get if you somehow stored your toothpaste in some queue-like structure where you only see 1 at a time, then yes, reaching the last would be helpful. But surely they are in a cupboard or drawer or something?
To know to start the sentinel toothpaste he must search around and prove it’s the only one left - by doing this he doesn’t need the sentinel anyway, as the searching process with yield only one, sentinel or not?
If it doesn’t need a search, e.g. they are in a line/all on easy view - then it’s obvious it’s the last even without a sentinel?
It helps if you are forgetful in a certain way. If you can remember to buy new toothpaste when you get the last tube out of the drawer, you don;t need a sentinel. But if you forget that the tube you just brought out is your last 5 minutes after you took it out, the sentinel reminds you.
Similarly with the emoji list. If can reliably remember which emoji is last, then each is equally as good as a sentinel. But if you can't, having one which is "special" will help remind you.
I assume OP's solution has thread-safety requirements. The toothpaste may be retrieved from long-term storage by another thread, and that thread may not be configured to check for termination conditions.
The benefit imo here is not that you realize you need more toothpaste only while you search for the last one, it's that while you're using it there's a clear signal to get more.
Is anyone is the AI/ML area finding success with anything other than conda, where installation of CUDA/CUDnn is required? Although I often have to pip install a lot of packages, I find conda's nvidia/pytorch/conda-forge channels are still by far the easiest way to get a deep learning stack up and running, and so I just stick with conda environments. I've tried poetry in the past but getting the NVidia deep learning stack up and running was really tough.
For anything related to CUDA/CuDNN, use one of NVIDIA base Docker images. Then whether you use Conda / Pip / Poetry / Pipenv does not matter much. Not at all a Conda fan myself and avoid it like the plague
What's surprising to me is that this isn't better known. The only reliable solution I've found is to go with the pytorch or deepstream images from NGC. Conda is probably a good idea for noobs who need Cuda installed for them on windows, but otherwise I find it an endless source of finicky issues, especially for unsavvy ML scientists who are looking for a silver bullet for package management.
10 years ago, « Data Science » work past the experimental stage was performed by SWE with a knack for applied maths. So investing in tooling to do things properly was a given.
Nowadays, most DS people only want to do ML at the experimental stage only and get lost when things get on the engineering side of things. But for their defense, nowadays the bare minimum skills require to do programming, containerization, CI/CD, etc. More experienced and swiss army knife SWE/MLE have to educate the willing.
It was already the same 10 years ago with MATLAB dudes not wanting to get dirty with C/C++/ASM SIMD. The history repeats itself, only at a faster pace
and it works. It used to not, but it's been fine for me for about a year now.
There's a very good chance I've installed cuda on my system before this though. And usually cudnn and some other packages because this is part of my standard install. And then I also never run into the issue where a package is looking for nvcc.
I have successfully transitioned an ML/AI team of seasoned researchers away from conda and to poetry. Some also use pyenv, I suspect a lot don't bother but may get bitten eventually.
It's definitely a learning curve, but it turns out every conda user has been bit by the irreproducible tendencies of conda quite often. Nobody uses the conda env file, they just start an env and pip install things into it. They don't realize the base env has stuff, too, and conda envs are hierarchical rather than isolated. I know it's possible to use conda in an isolated and reproducible way, but have yet to meet someone that does so.
So it hasn't been hard to pitch poetry to these folks, and while many complain about the learning curve they appreciate the outcomes.
We're a pytorch shop, and torch mostly just works with pip or poetry these days, as long as you skip the versions the torch maintainers mispackaged. We rarely need anything higher-level that only conda could install.
We really like having more than two dependency groups as this allows us to keep research and production in the same repository. main, dev, research. Then researchers contribute to the core library of a project and keep research and production using the same code for running and evaluating models.
So fast it finally made virtual environments usable for me.
But it's not (yet) a full replacement for conda, e.g. it won't install things outside of Python packages
Pyenv just worked for me. I am actually using Fedora Silverblue and have GCC and the CUDA SDK available only inside a toolbox container. Therefore, I have to enter that toolbox to install things like FlashAttention.
Have you tried https://pixi.sh/ ? It brings Cargo/NPM/Poetry like commands and lock files to the Conda ecosystem, and now can manage and lock PyPI dependencies alongside by using uv under the hood.
I haven't been using anything CUDA, but the scientific geospatial stack is often a similar mess to install, and it's been handling it really well.
How are you installing Pytorch with CUDA with Poetry? I stopped using Poetry because it wouldn't automatically get the CUDA version; instead, it would install the CPU version. I migrated to PDM, which does the right thing.
I stand corrected. I was familiar with the first option, which coupled the dependencies with the platform, whereas I wanted a CUDA version on Linux and a Metal version on macOS.
However, this works perfectly with Poetry 1.8 and Pytorch 2.2. I suppose the only problem is what PDM also does, where the lock file is platform-dependent. I'm not sure whether Poetry allows you to select a specific lock file, however.
was this before torch 2.0? With the very notable exceptions of a few mispackaged versions, torch now includes all the relevant Nvidia libs, and I haven't seen it grab the CPU version on a GPU box yet, though I'm not sure what it looks for.
A notable open issue in poetry is we can't currently specify one dependency on torch, and have it grab CPU version on some systems and GPU on others. Does PDM solve that?
I don't think PDM solves that directly. What I do is have different lock files for different platforms (e.g. Linux/CUDA and macOS/Metal), but pyproject.toml lists only "torch".
Lisa Su only joined AMD in 2012?