alcasa

That's not how this works. It really depends on how close you are to a position, where people might want to bribe you. Low provincial officials with direct ties to local land development might e.g. be able to take many more bribes than a highly ranked official in an office that is far removed from economic activity.

They should have used Figher Jet codenames instead. The MiG-15 one has a nice ring to it.

They forgot how to do pretraining.

I used it as a daily driver back in 2014/15 and it worked ok from what I remember.

Clinical research is in a weird spot, where you need both clinical experience and research experience. Getting the former already requires long hours and you are swamped with work. The latter is highly age gated and if you want to pursue research you often need to achieve specific milestones before a certain age, e.g. to be considered for tenure etc.

Isn't lisp getting advertisement every couple months here or so. After so many decades I don't believe the reason lisp doesn't have traction is just marketing.

Employer cost is not 2x, more like 1.2x, employer overhead is mostly insurance related stuff. We had salaray to employer cost tables at my previous job.

What true though is that after taxes you might just receive 60% of your total salary once you deduct taxes and insurances.

I fear people will just get used to it. Nobody gets tailored clothing anyhmore and people don't question that we have standardized sizes that don't really fit anyone properly. People commonly buy standardized furniture and rarely get something to a specific for their room. If cheaper software (I mean thats mostly what it is) gets the job done, we will probably just keep doing that, even if that means we lose something in the process.

Maybe controversial, but I believe a lot of OOP/Clean Code patterns are the software equivalent of corporate BS.

Frankly the most critical question is if they can really take shortcuts on DV etc, which are the main reasons nobody else tapes out new chips for every model. Note that their current architecture only allows some LORA-Adapter based fine-tuning, even a model with an updated cutoff date would require new masks etc. Which is kind of insane, but props to them if they can make it work.

From some announcements 2 years ago, it seems like they missed their initial schedule by a year, if that's indicative of anything.

For their hardware to make sense a couple of things would need to be true: 1. A model is good enough for a given usecase that there is no need to update/change it for 3-5 years. Note they need to redo their HW-Pipeline if even the weights change. 2. This application is also highly latency-sensitive and benefits from power efficiency. 3. That application is large enough in scale to warrant doing all this instead of running on last-gen hardware.

Maybe some edge-computing and non-civilian use-cases might fit that, but given the lifespan of models, I wonder if most companies wouldn't consider something like this too high-risk.

But maybe some non-text applications, like TTS, audio/video gen, might actually be a good fit.

Didn't know the L in Samuel L Jackson was for LeCun.

Fair enough. I feel like designing AI-proof take-homes is getting ever more futile. Given the questions need to be sufficiently low context to be human-doable in a short time and timespans for AI tasks increasing, I'm not sure take homes can actually serve any filtering function whatsoever, besides checking if applicants are willing to put in a minimal amount of effort.