industry-1

Claude takes over the programming of human computers, programmers are in a frenzy!

AI takes over human computers, the future is here! Claude controls computers autonomously and can do everything from programming to scientific research, representing an all-purpose API. on the other hand, OpenAI’s internal research and development of multi-intelligent body AI has been expedited and has already taken shape.

AI operating computers like humans has become the next frontier.
Two days ago, Anthropic evolution Claude 3.5 Sonnet stunned everyone by achieving autonomous control of the computer screen, moving the cursor, and completing tasks for the first time.

At the time, Anthropic’s head of developer relations said, “‘Computer Use’ is the universal API, and it represents the first step in a whole new paradigm of human-computer interaction”.

Netizens who got the beta test went crazy to experience this ‘superpower’.
Even the CEOs of startups are raving about it; AI intelligences have arrived, and you can now build AI armies to work for you.

At the same time Anthropic accelerated the layout of the future of the intelligent body, OpenAI also seems to feel a sense of urgency. The latest news that OpenAI is developing new products to automate complex software programming tasks.

Yesterday, OpenAI research scientist and the father of derby said at a TED event that he had lost confidence in building o1 and was forming a multi-intelligence team at OpenAI.

This means that a whole new race is about to begin, where AI is no longer just about dialog generation, but about execution and manipulation.

Claude takes over human computers, scientific research job search coding at the touch of a button
Developers who get their hands on Claude’s computer-using abilities simply can’t stop creating.
From complex coding tasks to in-depth research to collecting ‘bits and pieces’ of information, many amazing typical cases were born.
After all, it’s the first of its kind, so Alex Albert, Anthropic’s head of developer relations, went ahead and gave it a try.
Using Claude’s computer and the bash tool, he downloaded a randomized dataset online, then installed sklearn and trained a simple classifier on the dataset.
Finally, the classifier results were available on the web page.

All these processes were done in less than 5 minutes.
The time duration is 05:22

The hints used in it, he also contributed:
Go to https://data.gov, find an interesting recent dataset, and download it. Install sklearn with bash tool write a .py file to split the data into train (You may need to inspect the data and/or iterate if this goes poorly at first, but don’t get discouraged!) Come up with some way to visualize the results of your classifier in the browser.
One developer has already started asking Claude to help him do his own research.

Claude can do the “are you human or not” verification for you.

To get Claude to take control of your computer, all you need to do is:

pip install open-interpreterinterpreter –os

Search for YouTube videos and skip the ads.
Claude can do it all, and it’s all yours to do with the built-in ads.

Claude can also fill out job applications for laborers. The developer below has asked the AI to automatically apply for Anthropic jobs.

Evolution (left) and old (right) Claude 3.5 Sonnet builds impressively in ‘My World’.

How will the newly upgraded Claude 3.5 Sonnet affect the progress of the ‘multi-intelligent body society’?
Intelligent body research startup Altera Al has the answer, with a new model that is the biggest upgrade to prolonged autonomy. Our 25 intelligences collaborated in ‘My World’ to collect over 40% of the different items in 20 minutes.

Analytical tools
Incidentally, Claude also reintroduced the ability to write and run code ‘Analytics Tools’ today as a big benefit for 1024 developers.

This feature is now live in Claude.

Assuming that Claude is asked to draw a visual graph of the progress of a sales channel, it can autonomously analyze the data to write code and give the requested visualization.
You can then, in Artifacts, have a detailed view of the data for the segmented items.

The race for AI intelligences is on, Anthropic steals the show
While the computer-using tool isn’t perfect, it represents one of Anthropic’s visions in AI:
To make Claude human-like, reading screens to autonomously operate existing software and accomplish a variety of complex tasks.

The workings behind this capability are that Claude first takes a screenshot of the screen, determines what actions need to be taken, and then executes those lines of action. Then, another screenshot is taken to determine what should be done next.
Imagine the wealth of new opportunities that could open up if intelligences were able to view the contents of the screen directly without having to rely on assistive features, or AI software that looks at the underlying code.
For example, when you build a website, if the text in a button accidentally goes beyond the button boundary, the AI intelligence fixes the problem after seeing it directly, eliminating the need to view the underlying code backwards step.
To take another chestnut, current website producers, have been very clever to hide the HTML code of ads on the website.
This makes it more difficult for AI-based ad blocking software to analyze the code and determine what needs to be removed to remove the ads.
However, ‘computer-using intelligences’ that can see directly into the ads themselves will find this task much easier.
However, the technology poses drawbacks starting with the fact that ‘screenshot manipulation’ is too costly and the AI tends to assume that its manipulation has been successfully performed.
“By the time it acquires a new screenshot, it no longer knows where it is in the process.

On the other hand, there’s the issue of privacy.
Previously, companies have banned employees from using programming tools like ChatGPT and GitHub Copilot, fearing that they might accidentally leak proprietary information or code to model developers.
OpenAI is in a hurry, ramping up AI to end the year on a new one

Under the heavy pressure of the successive releases of its arch-rival Anthropic, OpenAI has actually long opened a new layout.
Remember a few days ago, Sam Altman suddenly bubbled, “next month is the second birthday of ChatGPT, what birthday gift should we send it?

At that time, a large wave of netizens wrote down their wish lists.
Just now, there was a report that went viral that OpenAI plans to, in December, unblock a new generation of big models, codenamed Orion. according to the revelation, Orion will be trained using data synthesized by o1 and will be released around the second anniversary of ChatGPT. But unlike GPT-4o and o1, it won’t initially go live through ChatGPT, but will first grant access to companies that work closely with OpenAI (like Microsoft) to make it easier for them to build their own products and features. However, the netizens have not been able to dream for long, Altman came out to dispel the rumors: all of them are FAKE NEWS!
In terms of software development, OpenAI is currently working on several products and features:
One part simplifies the process of developing with OpenAI’s AI in mainstream code editors like Microsoft Visual Studio Code;
Another part is aimed at handling more complex software development tasks.

Sources close to the matter revealed that the OpenAI product is capable of handling software engineering tasks that would otherwise require hours or even days of human time, as well as automatically writing and executing code for complex applications according to customer instructions.
However, the specific release time has not yet been determined.
After all, code development is one of the early application scenarios for OpenAI’s big language model, mainly because AI-generated code can be quickly verified for usability.
Starting in 2021, the Microsoft GitHub team launched AI Copilot using the OpenAI Big Model to provide programmers with real-time code suggestions.
This was followed by ChatGPT, which came out at the end of 22, offering an easier-to-use, free alternative that quickly became popular.
OpenAI then managed to convince millions of programmers to pay to use the ‘upgraded’ ChatGPT.
They could experience the upgraded version of LLM before GitHub Copilot and were able to handle all kinds of development tasks through conversational commands. As a result, reports say that these features have put OpenAI-related subscription products on track to reach about $3 billion in annual revenue.
In terms of intelligent body layout, OpenAI is forming a multi-intelligent body team internally, and may be leaning towards the intelligent body field next.

Some time ago, they released Swarm, a multi-intelligent body framework, which also sparked the attention of the AI community.

Internal research assistant
It has been revealed that OpenAI has developed an ‘internal research assistant’ that can help improve work efficiency and has been well received by researchers.
Among its features, it includes generating code for experiments related to AI models.
This internal tool appears to be a step towards developing systems that can autonomously conduct AI research – a capability that requires not only programming skills, but also the ability to come up with ideas and brainstorms for new experiments, among other things.
OpenAI’s leadership has publicly stated that this goal could be realized within the next few years.
In addition, people familiar with the matter have revealed that OpenAI is considering developing an upgraded version of Canvas, a tool that takes on Anthropic Artifacts.
It enables conversations with ChatGPT while collaborating in a new canvas that is interactive, whether programming or creating.

As far as code is concerned, in Canvas users are able to have the AI review code, fix bugs, etc. with a single click to help understand the existing code base and project type.
However, they also need to do the tedious task of copying and pasting the code into the chatbot.
What OpenAI is hoping for is the introduction of more generalized AI intelligences, similar to Anthropic’s release of ‘computer-using intelligences’ that can take over human computers to handle a wider range of tasks beyond code.
Currently, OpenAI has internally demonstrated a preliminary version of an AI Intelligent Body capable of completing tasks such as ordering food online through a user’s computer.