keskival

I let Opus 5.5 and GPT 6 Astra to finish my old hobby project.

  Location: Spain
  Remote: Yes
  Willing to relocate: No
  Technologies: AI, software engineering, healthcare, maps, supply chain, engineering, leadership, foundation models
  Résumé/CV: https://github.com/keskival/cv/raw/master/docs/cv_eng.pdf
  Email: tero.keskivalkama@gmail.com
  LinkedIn: https://www.linkedin.com/in/terokeskivalkama
  Google Scholar: https://scholar.google.com/citations?hl=en&user=l7qMBpkAAAAJ

I have programmed computers since I was 6 years old in 1988. I have over 25 years of professional experience in different roles in the trifecta of AI research/software engineering/leadership.

I was one of the people growing the deep learning revolution in Finland organizing Cybercom Machine Learning Guild and later AI Morning, and created one of the first open source implementations for Google Wavenet. I have won two IndustryHack competitions. I have authored quite a lot of patents in deep learning.

Most recently I have worked to teach cancer biology to machine intelligences to exceed human level in oncology.

I am open to all kinds of senior and leadership positions which benefit from a generalist.

  Location: Spain
  Remote: Yes
  Willing to relocate: No
  Technologies: AI, software engineering, healthcare, maps, supply chain, engineering, leadership
  Résumé/CV: https://github.com/keskival/cv/raw/master/docs/cv_eng.pdf
  Email: tero.keskivalkama@gmail.com
  LinkedIn: https://www.linkedin.com/in/terokeskivalkama
  Google Scholar: https://scholar.google.com/citations?hl=en&user=l7qMBpkAAAAJ

I have programmed computers since I was 6 years old in 1988. I have over 25 years of professional experience in different roles in the trifecta of AI research/software engineering/leadership.

I was one of the people growing the deep learning revolution in Finland organizing Cybercom Machine Learning Guild and later AI Morning, and created one of the first open source implementations for Google Wavenet. I have won two IndustryHack competitions. I have authored quite a lot of patents in deep learning.

Most recently I have worked to teach cancer biology to machine intelligences to exceed human level in oncology.

I am open to all kinds of senior and leadership positions which benefit from a generalist.

OkCupid used to work on the basis of high bandwidth of information and excellent matches.

This wasn't profitable enough though so they changed to the Tinder-style swiping which doesn't work. More engagement interactions means more opportunity to show ads. If you find people good matches they will just leave the app.

AI was never for making money. Money is for making AI.

The overall goal isn't to get wealthy, but for the wealthy to get the foot in the doorway to gain influence in the core of the automated economic military industrial scientific system that is going to replace money-based economics.

It is about autonomous robot armies.

The example is wrong:

"A simple poka-yoke example is demonstrated when a driver of the car equipped with a manual gearbox must press on the clutch pedal (a process step, therefore a poka-yoke) prior to starting an automobile."

You would typically put the car to a neutral gear before starting up the car, clutch isn't required.

Furthermore, if there was such a poka-yoke preventing start-up when the clutch isn't pressed, it would prevent the common safety procedure when a car doesn't start and is in a dangerous position, for example on rails or in the motorway. In such situations you would drive the car with the starter engine alone, putting the first gear on, release the clutch, and start the car, thus moving it forward slowly by the battery and the starter engine.

Ok, what if an email has "click this link if it was you who tried to log-in", or "if it wasn't you"?

Will Microsoft automatically authenticate malicious actors, or block yourself from services built with assumptions that the email client won't auto-click everything?

It's not really about flooring or rounding, but whether one thinks of time indices as ranges or moments.

Days, as the author points out, are though of with "flooring", but more accurately it could be said that a date is thought of as a range between the times belonging to the date.

Minutes one can consider as ranges or time indices. There the error comes, in switching the interpretation of a start of a duration to an actual estimate of a point of time index.

"I’m not sure, because OpenAI doesn’t deign to share gpt-4-base, nor to allow queries of gpt-4o in completion mode."

I would guess GPT-4o isn't first pre-trained and then instruct-tuned, but trained directly with refined instruction-following material.

This material probably contains way fewer chess games.

And also the cable between Lithuania and Sweden:

https://www.theguardian.com/world/2024/nov/18/telecoms-cable...

LLM based automated verification surely isn't something that easily works out of the box, but that doesn't mean ventures shouldn't try to work on it.

The purpose of capital is to make progress from where we are now.

Nope, just the code which sets up the agentic system and related prompts.

Here's how it sounds like:

https://youtu.be/yNi0bukYRnA?si=NnmGMBIkHRybmt0q

No, the root repo had an MIT license from Microsoft in it. And they have now changed it to Apache, and removed the Microsoft copyright clause.

They didn't fix the licenses. Microsoft VSCode was originally MIT licensed, which PearAI is now relicensing as Apache, and has removed the Microsoft copyright clause in their root repo.

Similarly, with their extension repo which they forked from Continue, they now have two LICENSE files, one from Continue, one from PearAI, both Apache. This could be argued to be almost correct, but it would be clearer if they did it correctly and had one LICENSE file there with both copyright clauses from both PearAI and Continue.

Additionally they don't seem to use any contributor agreements, so people who contribute to their repos will maintain their own personal copyrights. That means that the copyright clauses on the licenses are technically invalid as the repos contain works from various individuals who have not given their permission to license their works in any way.

Disregarding the legal aspects, what they are doing is morally clearly wrong as they are crudely exploiting the works of others without contributing back to them in any way.

I suppose in the real world such stars would collide in the center of the sphere and possibly form a black hole before achieving the required density approaching infinity, and also catapult stars out so that they leave the system by exceeding the escape velocity without encountering an elastic wall returning them to the system.

Are there digital scans though?

Edit:

It seems there's a lot of digitalized materials in UCL online archives, but I would be surprised if all documents were digitalized:

https://www.ucl.ac.uk/library/digital-collections/collection...

In these types of things, especially as the CEO was described as very intelligent, they should assume the cryptowallet is actually in his own control.

There is little reason to assume the CEO in fact didn't orchestrate this whole purported "scam".

By default the contributors also hold copyright and need to consent to the separate licensing.

However, this is typically solved by using a contributor licence agreement (CLA) where all contributors click through a form before submitting a PR where they declare that they own the copyright for the PR and they give a license for the organization to relicense the work and derivatives. Sometimes the whole copyright is transferred to the organization in these agreements.

I didn't check but according to some comments here the CLA in this case is already embedded into the AGPL license.

In principle this scheme guarantees that the original organization always has special rights over all of the open source community, as they can dual license all the derivative works.

The point of an OSS license in these kinds of situations is pretty simple:

They need to pick a license which is salty enough so that big corporations cannot use it, but have to buy a special license from the authors.

Meanwhile they can support and benefit from the open source community without bureaucracy, and get a maximal userbase to create content on the web about this stack, and have a maximal number of people who know this stack and want to use it.

AGPL is perfect for this.

It's not a never-ending treadmill because the past isn't being modified. It would be fine by me if they adapted the wording to be for example "we validate a method proposed by various authors before".

Of course I would appreciate a citation, but more than that I don't like false claims of novelty to become cemented into the body of scientific knowledge.

The authors are made aware of the project at the time of revising a pre-print, so not correcting the claims is immoral.

Edit:

The word "novel" in the context of scientific publication doesn't mean "new to the authors", it means "not previously published".

More information:

https://chatgpt.com/share/bf9e0de3-e05b-426c-b682-4ffa5f600f...

I'm not sure if there is obligation to cite anything. It's a question of ethics.

A claim of proposing a novel method should be corrected in a scientific publication if shown false. Otherwise the author is knowingly adding falsehoods into the body of scientific knowledge, and their institution is allowing it.

It is almost impossible to proactively find existing art conclusively. However, this wasn't my expectation in this case. I would just like them to correct the facts in their preprint, and add a citation which is missing. They refused to do that.

If you know a reference to an even more prior mention of meta-judges, please let me know. I am happy to add references to the repository.

Very typically authors write "to our knowledge novel method", which is fine, but in this case the authors cannot write even that if they know the method isn't novel now.

It's not illegal to omit citing existing work, like it isn't illegal to lie. It is highly unethical to knowingly claim to propose a novel approach which has been publicly proposed, and even implemented before though.

In effect this is erasing independent practitioners as inventors of methods from the history of science, and claiming false credit.

They fail to follow scientific ethics by refusing to cite pre-existing art, so that they can claim novelty for their methods.

How many times has this happened? The corporate labs are erasing independent contributors from scientific history, by falsifying it.

They fail to follow scientific ethics by refusing to cite pre-existing publications, so that they can claim novelty for their methods.

How many times has this happened? The corporate labs are erasing independent contributors from scientific history, by falsifying it.

Isn't that close to what instruct-tuned LLMs do when they are adviced to first think out aloud before jumping to answering? They are generating their own output to their scratchpad, the attended context, to support their thinking.

Doesn't the same thing apply to corporate software? Pass a LeetCode hard and bring a backpack full of bricks, and go for it?

Accounts aren't always faceless. They can have real names, long histories and connections to other profiles.

All points described are simple artifacts of tokenization.