euphetar

Honestly, a lame release of mediocre models.

I was most excited about the "frontier tuning." Like, it will actually watch you do stuff and learn to do it for you? That would be actually interesting.

But no, it's just a data labelling interface: https://learn.microsoft.com/en-us/microsoft-365/copilot/copi.... You have to provide the instruction and give feedback and there is a whole UI with hour-lonf wait between steps. So basically they want you to do the labelling to train a model, or at least that's how it looks from the outside

Also the mission statement of Humanist AI is the most boring, but tries to sound way too grand. Like "all the cool labs have a mission statement, so we should also have one" vibes

I wouldn't call it a benchmark since it's just one sample. They do highlight a real problem, though. Computer use is immature right now and far behind language agents

Try playing fruit ninja via text and llm toolcalls though

No link to edits or specific article. Disappointed. This shouldn't be front page as it contains no substance besides speculation

People don't like having their time wasted

Appreciate the effort, but this is very basic and nothing you need the source code to understand. I was expecting a deep dive into what specific decisions they made, but not how an loop of tool calls works

Nice

While I do not have much sympathy for Altman, the article is very low quality and contains zero analysis

Yeah, maybe on the surface chatbots turned out to be chatbots. But you have to be a poor journalist to stop your investigation of the issue at that and conclude AI is no big deal. Nuance, anyone?

the tidbit didn't work though, I just tried with GPT-4o

It's 100% clear why his I initiative was buried. The guy is a reckless egomaniac jerk. He is very transparent that he would do illegal stuff on a whim, target civillians, escalate conflict with other nations without even consulting a superior, and then reveal his face to the whole world, spilling secrets left and right, for a minute of internet publicity. And do a reddit AMA. And put "I am single LOL" there, showing just how much he can be trusted with leading a government cyberwarfare task force

What I read is that guy did a clever DDoS, which is mildly impressive, and decided to become a twitter celebrity/NK assassination victim. I am no hacker, but I expect real hackers to be laughing their asses off.

Please provide a feature to load data from notion export

The scary part is that perhaps there are in fact LLM generated bot threads, but they are not so easily busted

Auqui-fire?

Planet Farms | Onsite, Remote Hybrid | Portugal (Coimbra) | Full-time | ML Engineer, Data Engineer

We make vertical farms. We are looking for two people to join our data team.

For the Data Engineer we expect enough proficiency to help us build our DWH/Data Mesh infrastructure from scratch. We have a basic prototype for our data infra, but we are looking to make it more mature and be a good foundation for years to come. You will be supported by the rest of the data team, but we expect you to bring on board experience with DWH design and ETL processes. The stack is BigQuery, Google Cloud, Airflow, ScyllaDB, Postgres, Flask/FastAPI. Apply here: https://planet-farms.factorial.it/job_posting/34108

For the ML Engineer we are looking for a mix of computer vision and software engineering experience. The tasks are about improving the efficiency of factories by building ML-based tools. We as the data team own our solutions end-to-end, so we expect you to train the model, deploy the model build the required APIs to ship it and monitor it. The good news is that most of the infrastructure is setup and you will be the third ML engineer in the team, so you will not be alone. The stack is Pytorch, ClearML, Flask/FastAPI, Airflow, ONNX. Apply here: https://planet-farms.factorial.it/job_posting/machine-learni...

When applying please specify that you came from HN

Planet Farms | Data Engineer/Analyst | Office in Coimbra, Portugal We make vertical farms.

Hybrid office job, Coimbra, Portugal. Relocation, assistance of lawyers at the expense of the company.

The task is to found the data team and grow into its lead. At first it will be just you, but during the year it is planned to hire more people to the team.

It is necessary to lay the foundation of Data Mesh (current leading idea, up to discussion) and data culture in the company, which will allow scaling to dozens of factories.

In the beginning, you need to work a lot with your hands on the backend and infrastructure: raise the database, write the API and everything else.

Requirements: Data Engineering experience, Data Analysis experience, Good people skills. You will have the support of our team, but we need you to be the driver of the project.

Contact: boris.tseytlin@planetfarms.ag

Planet Farms | Data Engineer/Analyst | Office in Coimbra, Portugal

We make vertical farms.

Hybrid office job, Coimbra, Portugal. Relocation, assistance of lawyers at the expense of the company.

The task is to found the data team and grow into its lead. At first it will be just you, but during the year it is planned to hire more people to the team.

It is necessary to lay the foundation of Data Mesh (current leading idea, up to discussion) and data culture in the company, which will allow scaling to dozens of factories.

In the beginning, you need to work a lot with your hands on the backend and infrastructure: raise the database, write the API and everything else.

Requirements: Data Engineering experience, Data Analysis experience, Good people skills. You will have the support of our team, but we need you to be the driver of the project.

Contact: boris.tseytlin@planetfarms.ag

Assuming web app.

1. Python.

2. Flask. Pure server-rendered HTML, no fancy frontend. Avoid JS as long as I can, just use jinja templates. If my app is not going to work with good old HTML and CSS chances are it's not going to work at all.

3. Classless CSS framework. I like this: https://watercss.kognise.dev/.

4. Sqlite3 or Postgres. Prefer postgres because with docker it's 5 minutes to set up. I use SQLAlchemy from the start because it makes life easier. If you ever deploy and worry about real data you add alembic for migrations.

5. One docker compose for everything.

6. Cheap VPS for like $3 per month.

That's it. I would like to upgrade my MVP stack, but it's battle-tested and it works well. After the 9 to 5 I only have like 3 brain cells alive so I need to get stuff done as fast as possible, which means using what I know.

One thing I often reinvent and reimplement is OAuth. Definitely "Sign in with google" is the best way to add auth to your app, but it's such a pain to set up.

No way to assign probabilities? I prefer metaculus

Very shallow article. Would like to see a list of mentioned "recent breakthroughs" about using hashes in ML besides the retrieval applications, because this is genuinely interesting

So many people comment about how Snowden can't say things like this because he is in Russia. I wonder: how does it invalidate the facts? He links to Wikipedia. Maybe you don't want to believe Snowden, him being a traitor. Don't you believe Wikipedia? I don't see how Snowdens identity is even relevant to the subject.

I am currently also playing around with this. The best part is that for storage you don't need to store the reconstructed image, just the latent representation and the VAE decoder (which can do the reconstructing later). So you can store the image as relatively few numbers in a database. In my experiment I was able to compress a (512, 384, 3) RGB image to (48, 64, 4) floats. In terms of memory it was a 8x reduction.

However, on some images the artefacts are terrible. It does not work as a general-purpose lossy compressor unless you don't care about details.

The main obstacle is compute. The model is quite large, but hdds are cheap. The real problem is that reconstruction requires a GPU with lots of VRAM. Even with a GPU it's 15 seconds to reconstruct an image in Google Collab. You could do it on CPU, but then it's extremely slow. This is only viable if compute costs go down a lot.

My uninformed take for what it's worth.

Let's assume your projections are right and cannibalism is unavoidable like you say. You will never sell any of this technology. There is no need for it until there is a nuclear winter, so nobody will buy before it happens. After it happens nobody will buy anything. The only thing that could invest in something like this is government, but no government will commit political suicide by funding machine-assisted cannibalism.

Sounds like the same technology could be used to kill animals in a more ethical way. Perhaps you could market it as an ethical cow killing machine or something? Then you might even be able to continue secretly pursuing the cannibalism tech agenda, making sure the cow-killers can be easily repurposed into man-killers.

Anyways it sounds like you are an aspiring Bond villain, so I wish you good luck fighting spies and superheroes. When you inevitably trap one in one of these capsules don't turn around until they are 100% dead! Thats how they get you!

I have found the following strategy to work well: pick your battles. Some employers use test tasks to keep you busy so that you can't apply elsewhere. Other employers genuinely test your skills, but even among them there are some with unrealistic expectations. You can't be doing all the test tasks, period. But you can avoid most of them without walking away.

I am assuming an active job search where you have more than one potential employers.

If the test task is simple and you have the time: do it. If the test task is interesting and you really want into this company: do it perfectly. If the test task is simple, but you don't have the time: write an email explaining how you have a full time job with lots of interviews, tell them your CV clearly demonstrates you have enough experience, ask to forgo the test task. If the test task is hard and you don't have the time: explain how you would do the task, ask to forgo the task as above, walk away if they don't agree.

You would be surprised how often people will skip steps of the interview process if you ask nicely. This way I skipped tedious home assignments, psych tests and leetcode home assignment (not even hard, just told the company I don't want to rehearse dynamic programming with so many interviews lined up).

In my recent job search I had two good offers and incidentally both were the ones where I did the test tasks 150% because they were interesting. So do them if they are worth it.

Amazon has a UI problem. Unless you have a kindle it's extremely hard to buy a book and just get the god damn epub file. It's so much harder than pirating the book that I am not surprised. A few times I had to do it too: I would buy the book in the store and then go download it from a pirate website because amazon UI is too much pain

I botched the submission, please remove this post if you can

I believe the mods replaced the medium link with the github one. The github was the source for importing into medium, so I don't think it's proper. All the formulas and images got destroyed.

Could we please just share the original medium link? https://medium.com/toloka/reinforcement-learning-without-rew...

Rule of thumb: if you see a galaxy-brain style graphic next to anything "AI" or "Machine Learning" then the content is probably not worth your time.

This article seems to be an exception to the rule

It's indeed a very challenging task. I picked it because it's one of those where the lack of generalization is apparent. In fact, researchers study commonsense QA and similar tasks to find how we can reach generalization.

I agree, it's very hard to fidn the difference. Especially for GPT or BERT, where the training set is basically the whole internet.

It's a very good question about fur. I would suspect that it would correctly recognise all kinds of dog-like fur and sometimes fail on different furs, like bear furs. But in general NN's are very good at textures, so maybe it will just be good on all kinds of fur. One problem that might arise is that you will show it something fur-like but not really fur, and it will think it's fur.

I agree it's some kind of generalization. Here I don't have enough background to draw the line, perhaps a more theorically oriented person could, but I can't.

I guess the most important judge is that if you devise a benchmark like commonsense Q&A, a neural network fails it. Or how a Tesla will recognise a truck full of red stop signs as a real stop sign, while a "generalizing" thing like a human would definitely know that a core property of a stop sign is that it should be installed near a road. So there is a real problem.

It's repeating the same patterns it learned from text. Maybe with different specific objects, but still. One good way to test this is to ask it to count. It will break very soon. A person is able to build a rule in their head: "one apple is 1, two apples are 2, three apples are... 1+2 = 3".

Let's test GPT-2: https://huggingface.co/gpt2?text=One+apple+is+1%2C+two+apple...

Prompt: "One apple is 1, two apples are 2, three apples are" Model output: "One apple is 1, two apples are 2, three apples are 4, four apples are 5, six apples are 7, seven apples are 8, nine apples are 10, ten apples are 11 (for apples being the perfect length of life)."

Even if you use a special dataset to each it to count, it won't be able to count beyound the examples in the training set. So it's spewing plausible-sounding gibberish at you (i.e. approximating the training set distribution)

It doesn't generalize. Not in the sense that it can't give you a phrase that didn't exist in the training set. It can. But it can't give you a new kind of phrase, of a "kind" that didn't exist in the training set.

If that was true, a model trained on one commonsense Q&A dataset would be able to answer questions from another commonsense Q&A dataset without finetuning. But they can't. It's the same for every task you can find, but especially evident on commonsense reasoning benchmarks. At least last time I actively researched the question, I haven't found a single task where NNs definitely generalize.

When researchers dig in, they find that the neural network is learning wrong things. Word matching between answer and question, learning to model the annotator who asked a lot of questions because his Q&A answers are predictable, and stuff like this.

There was a great review of the problem, but I can't find it, so I will have to link to this article [1], which gives an overview of issues with current NLP models.

[1] https://thegradient.pub/frontiers-of-generalization-in-natur...

It's like this: no one is dependent, but it would be great to be able to support my relatives, they need it now.