Remote – o’reilly

Among the holidays, the end of the year projects, the next holidays and other hysteria, I have not found an article this month. So here there is a quick list of things that have surprised me recently.

Are we already virtual?

I am far from being the first person to find Notebook Incredible, and it certainly will not be the last. I did a simple experiment: I pointed it out in two of my recent publications, “Think Better” and “Henry Ford makes AI”. Both the summary and the suggested questions provided by Notebooklm were quite good: they went beyond simply commenting on the two pieces and entered the relationship between the two. But what surprised me was the podcast It generated: an eight -minute discussion between two synthetic people who sounded interested and committed. (Here is a description of some of the Techniques that Google puts to use For this to happen.) Was it 100% correct? No, but honestly, if a human summarized my articles, I would probably find some things I would complain about.

Being Google, after the initial experience, the user interface was more than a bit clumsy. When I wanted to return to the podcast a few days later, I had to play “to guess what to click” too much. (Suggestion: I suppose you should click on “Notebook Guide”? Why does the Podcast player not appear by default?) But that is really a very minor problem.

Models that use computers

Anthropic Computer use API Now it is available in Beta. Beta is right, there are clearly many things that are dangerous and easily abusive. But it is also very fun, and points towards a new direction for the development of AI.

In essence (and I can have the essence badly), the use of the computer allows you to tell Claude how to use a computer: browsers, editors, shells, anything that a user interface may have on a screen (and possibly more). Anthrope provides a demonstration as a Docker container, so he can run it safely. Once the container is running, it can give Claude a problem to solve; You will discover how to solve that problem and use the virtual Linux computer of the container to do the job. For example, you can ask you to complete a spreadsheet with data that collects from websites. Claude will do the whole click, copy and glue for you.

Is this revolutionary? My first reaction was “much, I can upload a file to GPT and use it to explore the website for me.” In principle, that is true, although ChatgPT does not allow web navigation and file loading in the same conversation. What is really new? Think about the monstrous message that you would need to read a spreadsheet, discover what data were missing, look for that data on the web and generate a new updated spreadsheet. It would not be simple. With the use of the computer, most of that complexity disappears.

Does it really disappear? We will discover as we advance. We are still in the stage in which hallucinations and bad behavior are cute instead of critics. It is easy for Claude to be deceived to interpret something on a random website as a warning. It will be a field day for immediate injection attacks. And I can imagine many improvements. The use of the computer currently works by taking screenshots and sending them to Claude, which calculates where to click. That seems incredibly uncomfortable, especially since many applications have accessibility possibilities that could make screen capture unnecessary.

For now, relax and breathe. Do not use the use of the computer at all, it is important to pay attention to the many warnings of Anthrope. But you must play with him and think about what it means. An automated frame to test web applications, Selenium ++? A tool to negotiate with online suppliers? We are much closer to a world full of agents where we ask a computer what to do and do it for us.

Could this be the end of CRM?

Something on the same line: Sam Lessin aware On Twitter (I will not call it x) on a very intelligent and useful trick. He exported many years of email, used GPT to extract key pieces and loaded it in a notebook (yes, again), which allows him to ask questions about his conversations during the last decade. Who did I talk to? Because? What are the issues we are talking about? That is all useful information.

Sam argues that this is the end of the structured software of customer relations management (CRM). I will not offer an opinion for investors or founders, but its process resonated with me immediately. I have worked with many potential authors and authors over the decades, and my email includes conversations with thousands of people. So, when I want to ask a question like “I want to understand more about the DDO; who should I talk to? My first step is to go to Gmail and start looking. Email is my CRM system; I have never used a CRM commercial product.

Unfortunately and ironically, Gmail’s search capacity is quite poor. Using it for contact management, although it can be operated, it is not pleasant. Can I ask Notebooklm? Absolutely.

CRM based on email could even be a good start of start, although it is difficult to imagine to succeed in the long term. There would not be much “pit” to protect a startup against larger companies, such as Google. I can easily imagine Google by building this type of search enabled for AI directly in Gmail. They already have all the data.

That’s all for this month. That was not so bad, maybe I should do this more frequently.