Google AI and DeepMind News and Discussions

[weatheriscool]

Google AI Supports Human Image Generation Again
Google paused human images in its Imagen model earlier this year following an online backlash.
By Ryan Whitwam August 29, 2024
https://www.extremetech.com/computing/g ... tion-again

Google has announced the rollout of a new version of its Imagen 3 AI image creator in Gemini. In addition to the usual spate of refinements and promises of prettier pictures, Google is once again allowing its model to generate images of people. This comes after it paused human generation in Imagen 3 in February after its initial release overemphasized diversity.

With the new Imagen 3 launch, Google promises better, more accurate depictions of your text prompts. The AI can again produce images of people, but there are still plenty of limitations there. The model will not make photorealistic images of identifiable people like celebrities and politicians. Likewise, it will block images that are "excessively gory, violent, or sexual."

[Vakanai]

Google AI Supports Human Image Generation Again
Google paused human images in its Imagen model earlier this year following an online backlash.
By Ryan Whitwam August 29, 2024
https://www.extremetech.com/computing/g ... tion-again

Google has announced the rollout of a new version of its Imagen 3 AI image creator in Gemini. In addition to the usual spate of refinements and promises of prettier pictures, Google is once again allowing its model to generate images of people. This comes after it paused human generation in Imagen 3 in February after its initial release overemphasized diversity.

With the new Imagen 3 launch, Google promises better, more accurate depictions of your text prompts. The AI can again produce images of people, but there are still plenty of limitations there. The model will not make photorealistic images of identifiable people like celebrities and politicians. Likewise, it will block images that are "excessively gory, violent, or sexual."

Is "overemphasized diversity" what they're calling all those pictures of black nazi soldiers?
Because the emphasis on diversity wasn't the problem, it was the complete inability for the model to get context.

[spryfusion]

[wjfox]

[raklian]

Congratulations to them! Very well deserved!

[Tadasuke]

This new and interesting 26 minutes long interview with Demis Hassabis by The Times about AI is worth listening to:



He says AGI in around 10 more years.

[firestar464]

Based on what people keep saying, AGI is either late 2024-early 2025, 2026, 2034, or never.

[raklian]

AGI is never late, nor is it early, it arrives precisely when it means to.

[funkervogt]

OpenAI has a financial incentive to declare that it has built an AGI. For a while, I've suspected that "the creation of the first AGI" will be a historical moment mired in controversy from the start, as it will fail to pass some tests of intelligence and will make mistakes uncharacteristic of an intelligent mind.

Oddly, that could be the key to getting out from under its contract with Microsoft. The contract contains a clause that says that if OpenAI builds artificial general intelligence, or A.G.I. — roughly speaking, a machine that matches the power of the human brain — Microsoft loses access to OpenAI’s technologies.

The clause was meant to ensure that a company like Microsoft did not misuse this machine of the future, but today, OpenAI executives see it as a path to a better contract, according to a person familiar with the company’s negotiations. Under the terms of the contract, the OpenAI board could decide when A.G.I. has arrived.

https://www.nytimes.com/2024/10/17/tech ... -deal.html

As I've predicted before, I doubt there will be a single moment in time when computers transform from "not intelligent" to "intelligent"; rather, they will proceed across a continuum of cognitive abilities. Depending on what metrics of intelligence you choose, you could assign any one of several different future dates as the "day that the first AGI was created."

By the same token, looking back on the history of human evolution, we can't say when the first hominid became truly "intelligent."
https://en.wikipedia.org/wiki/Human_evolution

[weatheriscool]

Google's Rumored 'Jarvis' AI Might Be Operating Your PC in December
Google's new AI could browse the web for you.
By Ryan Whitwam October 28, 2024
https://www.extremetech.com/computing/g ... n-december

Despite pouring billions of dollars into generative AI, Google has yet to make the case for this technology. Google is not alone in offering chatbots and image generators, but no players have managed to justify all the billions spent on AI accelerators despite the ballooning scale of models. Google's next AI model could be its shot to prove generative AI is worth the cost. Rumors suggest the new model will feature a "computer-using agent" that can automate research, shopping, and other tasks. You just tell the robot what to do and turn over the controls.

Jarvis, which is named after the AI in Marvel's Iron Man, will allegedly be able to control a computer in response to the user's commands. The mechanism by which it does this will probably sound a bit familiar. The model reportedly captures frequent screenshots, feeding them into a model that interprets the shots before implementing your will by tapping buttons, scrolling, and typing text. This is reminiscent of Microsoft's upcoming Recall feature for Windows, which captures screenshots to make your activity searchable.

[weatheriscool]

[firestar464]

(I was able to reproduce that last bit)

[firestar464]

[firestar464]

It's now a CBS article.

https://www.cbsnews.com/news/google-ai- ... lease-die/

[firestar464]

OpenAI response in 3, 2, 1...

[raklian]

[wjfox]

Google DeepMind predicts weather more accurately than leading system

Wed 4 Dec 2024 16.00 GMT

For those who keep an eye on the elements, the outlook is bright: researchers have built an artificial intelligence-based weather forecast that makes faster and more accurate predictions than the best system available today.

GenCast, an AI weather program from Google DeepMind, performed up to 20% better than the ENS forecast from the European Centre for Medium-Range Weather Forecasts (ECMWF), widely regarded as the world leader.

In the near term, GenCast is expected to support traditional forecasts rather than replace them, but even in an assistive capacity it could provide clarity around future cold blasts, heatwaves and high winds, and help energy companies predict how much power they will generate from windfarms.

In a head-to-head comparison, the program churned out more accurate forecasts than ENS on day-to-day weather and extreme events up to 15 days in advance, and was better at predicting the paths of destructive hurricanes and other tropical cyclones, including where they would make landfall.

“Outperforming ENS marks something of an inflection point in the advance of AI for weather prediction,” said Ilan Price, a research scientist at Google DeepMind. “At least in the short term, these models are going to accompany and be alongside existing, traditional approaches.”

https://www.theguardian.com/science/202 ... ing-system

[firestar464]

It remains the strongest even with stylecontrol

[weatheriscool]

Google Is Testing Gemini AI's 'Help Me Create' Functionality in Google Docs
Google's new AI tool for Docs automates document creation, integrating Workspace files for select testers.
By Devesh Beri December 9, 2024

Google Docs is getting a new feature called "Help me create," powered by Google's AI, Gemini. This tool, experimental for now, is made to assist users in generating formatted documents from scratch, adding to the already existing "Help me write" functionality that provides help with, as the name suggests, writing, rewriting, and editing text.

The "Help me create" option lets users start many types of documents, such as proposals, project trackers, blog posts, dinner menus, and more. As reported by 9to5Google, users can include content from their Google Workspace files by simply typing "@" followed by the file name. Google suggests sharing details like audience, purpose, and goals to get the best results from the tool.

https://www.extremetech.com/computing/g ... oogle-docs

[weatheriscool]

Google Announces Gemini 2.0 to Kickstart an Era of 'AI Agents'
The first version of Gemini 2.0 is available to use now.
By Ryan Whitwam December 11, 2024

https://www.extremetech.com/computing/g ... -ai-agents
Time sure flies when you're being bombarded with AI models in every technology product in the world. It has been just a year since Google revealed Gemini 1.0, and it's celebrating by announcing the latest iteration, Gemini 2.0. Like the current v1.5 models, Google plans to have multiple flavors of Gemini 2.0, but it's starting with the smaller, more efficient Gemini 2.0 Flash, and you can try it now.

Gemini 2.0 was built for what Google insists upon calling the "agentic era," which will be marked by AI models (agents) that can act autonomously on your behalf. Gemini 2.0 will power some of Google's previously teased AI projects, but it's starting relatively modestly with Gemini 2.0 Flash. While Google is making some big promises for what Gemini 2.0 will do, its current implementation is optimized for chats.

Demis Hassabis of Google's DeepMind AI division says Gemini 2.0 Flash outperforms the larger Gemini 1.5 Pro on multiple metrics. At the same time, it's twice as fast. That means you won't have to wait as long for text to appear, and Google saves some computing resources. It supports multimodal input as well as output. That means you can feed it images, video, or audio like the 1.5 models, but it can also output multiple data types. Currently, that's limited to text and images (via Imagen 3), but it could include more data types later.