April 07 2023, 19:12

I am currently finishing a 1000-page work by mathematician Stephen Wolfram, “New Kind of Science”. I will write separately about this book, it is wildly interesting, and very different from anything I read before. I bought this tome — it is almost 10 cm thick, but the book is freely available online.

For today — just a fun fact — the author just really loves to mention the main idea of the book almost on every 20th page, almost with the same words. Well, okay, it’s relevant everywhere, but still. I quickly pulled out some excerpts.

The main idea is that the complexity of the world around us can be explained by simple rules, which are applied by a straightforward algorithm. Changes in the initial setup can reflect strongly in the final complexity — for example, in shapes or patterns.

April 07 2023, 11:14

How easy it has turned out to be to write articles in English. A translator plus editing took 10 minutes. Plus another 3 minutes to create an illustration in DiffusionBee.

April 07 2023, 10:34

Right now, ChatGPT is being used to create resumes, recommendation letters, and cover letters, and theoretically, AI could soon replace HR entirely, aiding in conducting interviews. This is already feasible, and if a trained version is prepared, it will be truly spectacular.

Remember, about ten years ago there was Akinator (and still is), which could guess the “character” based solely on responses to questions. It used a decision tree algorithm, the same as machine learning. In that algorithm, the connections between concepts were simple, but in what is now used in ChatGPT, they are exponentially more complex. So when you interview someone, you basically have a very limited number of questions (say, twenty) to understand their qualifications and what kind of person they are. Being able to structure a dialogue in such a way to “guess the hero” more quickly is a skill in itself, not everyone possesses.

This is exactly where AI could be extremely helpful. It might be odd to hand over interviewing entirely to AI, but making AI a tool for the interviewer is very plausible. Along with voice recognition and analysis of responses, this tool could provide recommendations on what to ask, and identify where there are clear “leaks”. It can take into account details from the resume and new information garnered during the interview, as well as results from previous interviews with other people in the same company.

Such interviews could train a brain on a corporate level or one big brain common to all companies in the market. The second path assumes that some information will flow into a central system, a big “brain”. This might meet resistance at various levels, but if the efficiency of recruiting good employees skyrockets, businesses and governments might turn a blind eye eventually.

Moving forward, developments could go in two directions. The first is this unified big “brain” will know specific people. If you arrive at a new company, the AI there already knows not to ask certain questions because they were asked just last week in another company. Even if AI does not share this with the company, it may decide that this knowledge correlates with the job description and informal expectations of company leaders, and simply not waste time repeating this question. This path will face rejection from many people, but from a business standpoint, it’s very beneficial.

Essentially, everything may lead to a social rating system like Nosedive from Black Mirror, only there won’t be just one number. There will be billions of numbers, understandable to AI, from which it can make recommendations about a person’s suitability for a specific task.

The second path is if such interviews train a system that is not tied to specific individuals. But like in Akinator – based on the analysis of resumes and past interviews, the likelihood that a candidate will respond in a certain way to a particular question might be so high that AI simply won’t ask it. Especially if the response to that question is already considered in AI’s assessment of whether the candidate fits the job description or not.

In this case, the social rating is not tied to the individual, but it is tied to the “cloud” of that person’s properties, their history, and the traits of their personality, which all together pinpoint them quite accurately. Yes, they won’t have a direct link to the specific individual, but essentially the outcome will be the same as in the previous case. The recommendations will be the same.

I think that integrating AI into the hiring process will change significantly more than just spotting certain specialties. I believe such optimization will make the contrast between those who are working and those who are job hunting much stronger. It will become extremely difficult to land a good job just because you got “lucky” and the interviewers accidentally didn’t touch on a huge range of topics that the candidate would likely have failed. And the main thing—what objections could there be, since AI only helps in hiring those suitable for the job, not just anyone? And the government will ask the public: Or did you want it otherwise? Do you want doctors and teachers in schools to be hired just because they were “lucky” enough to pass an interview?

What do you think?

April 06 2023, 18:26

Yesterday until four in the morning, I was tinkering with this thing. It works! ChatGPT, Google AIY, can talk about the weather and forecasts (still slightly unreliable). And about everything under the sun. Keeps the session. Microphone is temporarily external. Voice recognition – openai whisper. Synthesis – Google TTS. Dialogues – openai GPT 3.5. Weather – tomorrow.io Inside – raspberry pi. I made a wakeword, but decided not to use it. It consumes the voice recognition API inefficiently.