What type of coding are you doing? Did you locally roll your own coding assistant with a local model of DeepSeek or are you prompting via the web?

After graduation I got a job at a top 10 nationally ranked engineering firm which has a very strong reputation. My coworkers were the top of their class coming from much better engineering schools. Compared to them my education was better as judged by the number of calculation mistakes in my work vs theirs. I also had a better understanding of some core concepts. So comparatively I do not regret my degree given I landed where they did, with fewer student loans and a comparable education but I definitely got lucky.

Overall I would say my studies prepared me very well for certain things like working hard, preserving through difficult problems, learning how to be more analytical etc but it really didn’t prepare me for basic usage of the tools, learning the building code or working professionally with a manager or on a real engineering team as opposed to a team of students working on a BS design project in some engineering class where the professors don’t really grade the final output.

I think your experience is similar to mine. University teaches a “foundation” level of basic understanding and leaves out a lot of the basic professional skills or when they do try to “simulate” real work it is extremely phoned in. At first I did feel some amount of regret that my degree didn’t “fully” prepare me for professional working life but I think that’s just a result of dealing with “gaps” in development since we can’t possibly learn everything in our degree. If I had gotten internships during my undergrad degree then I think I would have probably had a more complete education with less buyers remorse.

As universities become more market as opposed to academic dominated I think this dynamic might invert.

The real regret came when I saw my paycheck and when I heard what software engineers make. I later switched into software and have never looked back.

If you were of similar size then yes. Basically being smaller really sucks in a fight. So if you can reasonably deflect the Rocks wrestling & grappling offensive you could rope a dope him causing him to gas out and then win. Higher weight MMA fighters & boxers use this tactic all the time. They let their opponent “punch themselves out” and then get off their own offense.

Comments URL: https://news.ycombinator.com/item?id=37390733

Points: 2

# Comments: 1

Company B isn’t just looking at company A’s “success”. Company B’s levels of attrition are increasing because its employees are going to company A. So company B has no other option than to open the hiring floodgates or else they can’t meet bare bones maintance tasks let alone growth.

Not your comment but most of the comments on this thread are whacky. It’s like people completely forgot how hot the software market was during the pandemic. Anecdotal but I had 2 friends trying to break into the software game at the same time. That didn't happen during the pre 2019 boom so who knows what that means lol

https://openai.com/blog/triton/

https://triton-lang.org/master/index.html

Maybe this is a naive criticism but if the goal of the model is to "emulate the correct reasoning function" by which is implied that it learns prepositional logic rules then shouldn't the model have some architecture that tries to model generic logic rules and then using that score the input data and predict the result?

Currently it seems like the model is simply scoring each predicate since each of it's reasoning layers performs one step of forward chaining, adding some predicates to the Proved Facts.

My naive (and way undercooked) way of conducting this experiment would have been to try to use a GAN where the generator model comes up with new predicates and then the discriminator model that tries to classify them as real/fake. I would then try to train an MLP on top of the generator to classify the result and swap out the MLP depending on the sampling method basically the generator model becomes the pre-trained thing we care about to see if it can generalize.

Another nitpick is that the authors claim that BERT has enough "capacity" to solve SimpleLogic BUT this isn't actually what they want to achieve since solving != learning reasoning. So it feels like a bait and switch since IMO if a model has the capacity for something then it has some architectural aspect to it that can be used to show that it can achieve a smaller version of what you want and they didn't prove BERT can learn ANY prepositional logic rule.

I think for Paul Graham it’s a different story since he talks to other influencers and occasionally goes viral. That sort of feedback loop well that’s quite addictive. There’s a reason why there’s no obvious Twitter competitor.