AI Eye – Cointelegraph Magazine

ChatGPT is lazy

Recently, more and more users have noticed that instead of doing the work, ChatGPT appears indifferent or provides partial answers and instructs the user to complete the task themselves. To me, that is a clear sign that human-level artificial intelligence has been achieved.

Many suspect that OpenAI nerfed it to reduce the system’s exorbitant costs, but OpenAI says otherwise.

Meanwhile, users bribed the AI ​​with tips for better answers and impressed it in prompts about how important the best possible answer was.

Google’s Gemini video was unreal.

As we now know, Google’s “amazing” and “unrealistic” product video for the Gemini Ultra was fake. But does the trick matter?

The video shows a man having a natural conversation with an AI, which recognizes that he is drawing a duck. Gemini Ultra can figure out which cup the ball is under during a shell game and can even figure out when someone is playing rock-paper-scissors.

But in reality, there was no video or audio conversation. The AI ​​received messages via text and simply displayed still images.

But Google says the prompts and output were real and that Gemini can actually recognize pictures of ducks.

This seemed like a huge breakthrough, as duck recognition had been unknown to scientists for a long time. But as it happens, ChatGPT can also recognize ducks.

Oriol Vinyals, vice president of research at Google DeepMind, said the video was just a suggestion.

“This video shows what a multimodal user experience built with Gemini can look like. We created this piece to inspire developers.”

Gemini’s test results showed that it outperformed GPT-4 in seven out of eight benchmarks. This sounds good until you realize that GPT-4 was completed a year ago, giving OpenAI a 12-month head start on GPT-5.

Fetch.AI = Google of AI agents?

AI Eye met this week with Humayun Sheikh, founder of Fetch.ai and former commercial director at DeepMind (now Google DeepMind). Sheikh said the company was “doing something similar to GPT 13 years ago.” This means that Google has deliberately decided not to release LLM until ChatGPT forces it to do so.

“I was very surprised that it came out quite late and gave OpenAI a chance. But I think they were ready a while ago,” he said.

“The problem is that if you start using this technology and neglect it, it starts to eat away at your own business.” He added, speculating that AI could undermine the profits of Google’s lucrative search business.

That said, Sheikh believes Google is still winning the AI ​​arms race.

“Google has Maps, Google has Ads, Google has Business and all the features you need to integrate with a set of applications,” he says. “So I think there’s a problem with OpenAI. We’re going to have to change drastically for OpenAI to be viable.”

Fetch is in the business of helping users create AI agents, with 100,000 of them running to date. Bots can help with travel bookings, EV charging, and IoT device management.

When you give an AI agent a goal, the AI ​​agent creates multiple subtasks and performs them autonomously until the goal is achieved. Sheikh says blockchain is a natural fit for coordinating agents and recording performance.

“Interacting with multiple entities, interacting with multiple pieces of code, interacting with multiple machine learning AI algorithms requires a new framework. It can’t run on existing Web 2.0 frameworks.”

“The coordination of these operations has to happen somewhere,” he says. “And these micro-tasks require the monetization that blockchain provides.”

Peaq Blockchain, Fetch.ai, and Bosch have unveiled the Bosch Network on which tokens are exchanged (DePIN).

For example, you could ask an AI agent to check a car’s tire pressure sensors, as well as sensors that record weather conditions, and find out if the tire pressure is too low or too high or if new tires are needed. That way, they can make an appointment for you at a tire dealer.

Alternatively, you can ask an AI agent to perform a sentiment analysis on a particular stock and buy that stock if the analysis is favorable.

Sheikh said Fetch plans to be a search engine for agents, “and it will also be a self-assembling engine that ties all of this together without the need for coding or integration.”

All killer, no filler AI News

— Meta launched 20 new AI features on its social media platform to power search, advertising, and business messaging. Free image generators are what people love.

— Alibaba’s Boffins scrapes TikTok influencer videos and can now generate photos of people dancing from still photos some pose guide.

— Professor Ethan Mollick compared and contrasted all LLMs and said the best LLM to use is GPT-4. Studies show that they actually improve job performance, generate better ideas than most humans, look smarter and have the most features than their competitors.

— McDonald’s is building an AI chatbot called Ask Pickles, trained on data from 50,000 restaurants. Franchisees and employees can now use AI to understand exactly how to do everything in a more McDonald’s way.

— Amazon currently has about 750,000 robots working alongside 1.5 million employees.

— Grok user Jax Winterbourne was surprised when the AI ​​refused to answer his questions and cited OpenAI policies as justification. Some users believe that Grok is nothing more than a front-end for ChatGPT, but Xai said that it picked up some of ChatGPT’s results from its training data.

Tweets of the Week

A new perspective has emerged. “The only thing an LLM does is hallucinate. The trick is to match the hallucination with reality.”

subscribe

The most interesting read on blockchain. Delivered once a week.

Andrew Fenton

Andrew Fenton, based in Melbourne, is a journalist and editor covering cryptocurrency and blockchain. He has worked as a national entertainment writer for News Corp Australia, a film journalist for SA Weekend and The Melbourne Weekly.

Follow the author @andrewfenton