Warning: This is AI generated, probably a low end model as some of the content is outright nonsense eg:
"""
concept of MoE is quite prevalent (refer Outrageously Large Neural Networks: the Sparsely-Gated Mixture-of-Experts Layer), with Langchain’s high-level implementation of an LLMRouterChain, and notable low-level integrated examples
"""
At first, I was confused because I didn't get why they would call it "hand-picked" when literally every single "article" on that site is attributed to the site itself.
And then I clicked a random page and, well, it's slop:
>By amplifying the signal from minority class data and leveraging the diversity of models, these methods enhance prediction accuracy and fairness across all classes. When paired with complementary techniques such as resampling, adjusting class weights, or generating synthetic data, ensemble methods can yield even more robust results in handling imbalanced datasets.
The whole article seems to hinge on the idea that cursor is unsustainable and a large driver of revenue for AI companies. It seems to think that if cursor dies, so does the revenue.
I don't see that- cursor will die by losing market share, not by the death of the market. Agentic coding as a market will continue to grow and if Claude remains competitive then Anthropic will do just fine.
if im a below average web developer of 15 years and wanted to transition to entry level ML/AI engineer of any sort, how much time would you guess it would take to become a competitive candidate if i worked at it full time? i'm intelligent but just completely declined to apply myself more to web dev, which is why i sucked at it.
i've been unemployed for two years and it's hard to find anything i can do with my background that isn't "software engineer". entry level stuff for adjacent things like project or product management simply aren't available. exploring options like data science seems to show that area is also extremely competitive and the job market is terrible.
i really do not want to grind leetcode for the sake of yet another job plumbing CRUD apps together, but i could see myself learning a new domain and following through with that because the end result would be worth it/interesting
i think my hope is that interviews for ML engineering would focus on domain knowledge specific to ML and not intricate questions about using typescript and react, which is the sort of stuff i simply cannot bring myself to care about enough to memorize well enough to discuss well in an interview
The premise is a basic understanding of matrices and stats. Then go through a course on machine learning (supervized, unsupervised, deep and reinforcement learning). That’s for the theory. Easy if you know your math and python.
The practical side is endless tweaking of the data and the models. And keeping yourself informed about new models and techniques (throug scientific papers)
thanks! i think i understand the nature of the job okay, i think its more my ability to realistically get there in 6-12 months is what i'm concerned about
Kudos to the author, a very very nice overview of things!
I was just at a big conference where there was now a lot of AI talk and papers etc and, as that hasn't been my area, I've been catching up. Have been hearing so much about GNN and ensembles and things that I haven't had to think about before. So now I'm going through this and looking up new-to-me terms. Sweet.
Now it's just to work out how to save it as an epub to browse on the beach.
It's obviously machine-generated from the linked source materials, and while I'm grateful for the bibliography, I wish I didn't have to click through a million pages to get from the blogspam to the actual content.
Warning: This is AI generated, probably a low end model as some of the content is outright nonsense eg: """ concept of MoE is quite prevalent (refer Outrageously Large Neural Networks: the Sparsely-Gated Mixture-of-Experts Layer), with Langchain’s high-level implementation of an LLMRouterChain, and notable low-level integrated examples """
Is it possible to label/tag these submissions as containing content that is AI-generated? I think the HN community would appreciate that
Yes, it would be appreciated.
No, it is not possible.
The paper itself is fairly popular, with several thousand citations.
Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, Jeff Dean
https://arxiv.org/abs/1701.06538
At first, I was confused because I didn't get why they would call it "hand-picked" when literally every single "article" on that site is attributed to the site itself.
And then I clicked a random page and, well, it's slop:
>By amplifying the signal from minority class data and leveraging the diversity of models, these methods enhance prediction accuracy and fairness across all classes. When paired with complementary techniques such as resampling, adjusting class weights, or generating synthetic data, ensemble methods can yield even more robust results in handling imbalanced datasets.
[dead]
https://www.wheresyoured.at/ai-is-a-money-trap/
The whole article seems to hinge on the idea that cursor is unsustainable and a large driver of revenue for AI companies. It seems to think that if cursor dies, so does the revenue.
I don't see that- cursor will die by losing market share, not by the death of the market. Agentic coding as a market will continue to grow and if Claude remains competitive then Anthropic will do just fine.
The author doesn't understand how much money is in the it industry and free market.
It doesn't matter if companies next to Google, Ms, meta make it.
Google does ai sustainable enough alone.
if im a below average web developer of 15 years and wanted to transition to entry level ML/AI engineer of any sort, how much time would you guess it would take to become a competitive candidate if i worked at it full time? i'm intelligent but just completely declined to apply myself more to web dev, which is why i sucked at it.
i've been unemployed for two years and it's hard to find anything i can do with my background that isn't "software engineer". entry level stuff for adjacent things like project or product management simply aren't available. exploring options like data science seems to show that area is also extremely competitive and the job market is terrible.
i really do not want to grind leetcode for the sake of yet another job plumbing CRUD apps together, but i could see myself learning a new domain and following through with that because the end result would be worth it/interesting
AI/ML dev is just boring old SWE but now you've got some new ingredients in the mix.
i think my hope is that interviews for ML engineering would focus on domain knowledge specific to ML and not intricate questions about using typescript and react, which is the sort of stuff i simply cannot bring myself to care about enough to memorize well enough to discuss well in an interview
The premise is a basic understanding of matrices and stats. Then go through a course on machine learning (supervized, unsupervised, deep and reinforcement learning). That’s for the theory. Easy if you know your math and python.
The practical side is endless tweaking of the data and the models. And keeping yourself informed about new models and techniques (throug scientific papers)
thanks! i think i understand the nature of the job okay, i think its more my ability to realistically get there in 6-12 months is what i'm concerned about
Exploring Kaggle and participating in ML competitions is an excellent way of getting hands on experience.
I was curious what the AI would include if promoted to create a similar list. I prefer the human version.
Ref: https://chatgpt.com/s/t_689a00f83f7c8191b70d07912a092f86
[dead]
Surprised to see something on hackernews that isn't an AI picked selection of articles :-)
/s
Kudos to the author, a very very nice overview of things!
I was just at a big conference where there was now a lot of AI talk and papers etc and, as that hasn't been my area, I've been catching up. Have been hearing so much about GNN and ensembles and things that I haven't had to think about before. So now I'm going through this and looking up new-to-me terms. Sweet.
Now it's just to work out how to save it as an epub to browse on the beach.
It's obviously machine-generated from the linked source materials, and while I'm grateful for the bibliography, I wish I didn't have to click through a million pages to get from the blogspam to the actual content.