UPP News: What's new in the AI field? (Part 1)

Date: Nov 23, 2023.

Bard’s YouTube extension can now handle complex queries about specific video content, like recipe quantities and instruction summaries.

Bard, Google’s AI chatbot, is getting a handy upgrade so it can analyze individual videos to surface specific information for you — like key points or recipe ingredients — without ever pressing play.

To enable the YouTube extension, all you need to do is go to the Bard homepage and tap on the puzzle icon in the top right corner. Once you do that, you will see multiple options. Just turn on the toggle in front of YouTube and you are good to go. You can then ask for the summary of any YouTube video by simply copy-pasting the link on the Bard homepage.

In a trial asking for the summarization of America's Test Kitchen's recipe for an Espresso Martini, Bard got all of the critical bits right in summing up the video: the ingredients and measurements are all accurate, and the instructions are correct.

The feature, however, may raise concerns among YouTube video creators by encouraging viewers to bypass the traditional process of watching videos. This means they skip the steps of viewing preroll ads, engaging with the channel's recommended videos, and potentially generating revenue for the content creator. While this might be convenient for viewers, it could pose challenges to the financial interests of video publishers, such as reduced ad revenue and missed opportunities for engagement with their other content.

Date: Nov 22, 2023

Sam Altman is to return as chief executive of OpenAI after the ChatGPT developer said it had “reached an agreement in principle” for his reinstatement.

Open AI was announced after days of internal turmoil after Altman’s surprise sacking on Friday.

The company wrote on the X platform: “We have reached an agreement in principle for Sam Altman to return to OpenAI as CEO with a new initial board of Bret Taylor (chair), Larry Summers, and Adam D’Angelo. We are collaborating to figure out the details. Thank you so much for your patience through this.” Under the agreement, an overhaul of OpenAI's governance, including removing three out of six board members,d was to be performed as a precondition for his return.

Previously, Sam Altman was fired by the board of directors, saying that it no longer had confidence in the CEO because “he was not consistently candid in his communications with the board, hindering its ability to exercise its responsibilities.” No examples of that alleged behaviour were provided. The four directors, representing a majority of the six-person board, also kicked OpenAI president and chairman Greg Brockman off the board. Brockman quickly resigned.

Altman posted on X soon after: “I love OpenAI, and everything I’ve done over the past few days has served to keep this team and its mission together.” He added: “I’m looking forward to returning to OpenAI.”

Altman’s return will make him OpenAI’s fourth CEO in five days. He was initially replaced by Mira Murati, the company’s chief technology officer, as interim CEO, before the board hired Emmett Shear, the co-founder of video game-streaming platform Twitch as a temporary successor.

Date: Nov 22, 2023.

Amid the chaos that has engulfed ChatGPT creators OpenAI over the last few days – which has led to CEO Sam Altman being fired and then quickly reinstated to his position – one of its direct competitors, San Francisco-based AI startup Anthropic, has upgraded the language model powering its chatbot, Claude AI.

Claude 2.1 delivers “advancements in key capabilities for enterprises” and now includes a 200K context window, which is bigger than the context window offered by GPT-4 Turbo and the largest context window offered by a publicly available AI tool.

Anthropic says that Claude 2.1 will be able to “summarize, perform Q&A, forecast trends, compare and contrast multiple documents, and much more”.

Another big change to Claude is a decrease in hallucinations – Anthropic says that Claude 2.1 is half as likely to serve users false statements as its predecessor, Claude 2.0.

Significant improvements have also been made to the chatbot’s ability to respond with correct answers to queries related to long-form documents, such as legal filings and financial reports.

Users can sign up and use Claude for free via Anthropic's website, where they can also view the price of token bundles to talk to the chatbot. The full 200K token capacity remains exclusive to paying Claude Pro subscribers for now. Free users will continue to be limited to Claude 2.0’s 100K tokens.

With its substantial context expansion and rigorous accuracy improvements, Anthropic’s latest offering signals its determination to compete head-to-head with leading models like GPT-4.

Date: Nov 17, 2023.

Microsoft announced its latest contribution to the artificial intelligence race: software that can generate new avatars and voices or replicate the existing appearance and speech of a user.

Announced at Microsoft Ignite 2023, Azure AI Speech is trained with human images and allows users to input a script that can then be “read” aloud by a photorealistic avatar created with artificial intelligence. Microsoft said in a blog post that the tool could be used to build “conversational agents, virtual assistants, chatbots and more”. The post reads: “Customers can choose either a prebuilt or a custom neural voice for their avatar. If the same person’s voice and likeness are used for both the custom neural voice and the custom text-to-speech avatar, the avatar will closely resemble that person."

The announcement quickly elicited criticism that Microsoft had launched a “deepfakes creator” – which would more easily allow a person’s likeness to be replicated and made to say and do things the person has not said or done. In a statement, the company pushed back on the criticism, saying the customized avatars are now a “limited access” tool for which customers have to apply and be approved by Microsoft. “With safeguards in place, we help limit potential risks and empower customers to infuse advanced voice and speech capabilities into their AI applications transparently and safely,” Sarah Bird of Microsoft’s responsible AI engineering division said in a statement.

The text-to-speech avatar maker is the latest tool as major tech firms have raced to capitalize on the artificial intelligence boom of recent years.

Date: Nov 10, 2023.

Humane has launched its AI Pin, one of the world’s most hyped tech products, and it has immediately made a public mistake.

The system is intended to be attached to clothing and then makes use of a range of microphones, speakers and a display that can shine onto its owner's hand to give information. Users can just press the AI Pin and speak into the air, which will then allow the computer to access the internet and show an answer.

During its reveal event, the pin was asked "When is the next eclipse, and where is the best place to see it?”. The AI Pin answered that the best place to view the next total solar eclipse, in April 2024, would be Exmouth in Australia or East Timor.

However, next year’s solar eclipse will in fact be visible in North America. It will not be at all visible in Australia, and can only be seen in Mexico, the US and Canada.

The system may have made the mistake because a solar eclipse earlier this year was in fact best viewed from Exmouth and East Timor. That eclipse, in April, brought widespread coverage to the small Australian town – and that coverage was presumably used to train the AI system that answered the question.

At the time, many noted that the error highlighted a central error with large language models. The systems tend to “hallucinate” – or confidently state falsehoods – and have no real way of being able to check whether the information they are given is true.

Date: Nov 8, 2023.

Not one to be left behind in the race for AI, South Korean electronics company Samsung has just unveiled its own generative artificial intelligence (AI) model, Samsung Gauss, an unusually named AI model that responds to user-generated prompts in a similar way to ChatGPT.

The new model will be incorporated into Samsung's Galaxy S24 handset in the first half of 2024, in an effort to improve its product offering and become less reliant on large language models (LLMs) provided by competitors.

Samsung Gauss is split into three divisions: Samsung Gauss Language, Samsung Gauss Code, and Samsung Gauss Image. As the name of these functions suggests, the AI tool is capable of generating text, code, and image responses, offering versatile applications to its users.

Samsung Gauss Language aims to streamline productivity in a number of different ways, including helping users write and edit emails, summarize documents, and translate languages.

Its coding function, Samsung Gauss Code, is primarily designed to help developers write code quickly, by supporting “code description and test case generation through an interactive interface.”Lastly, Samsung Gauss Image can create and edit images, similar to the OpenAI image generator DALL-E.

Samsung Gauss hopes to have an edge over rivals when it comes to privacy. Unlike most other AI models, the tool is an “on-device AI.” This means it will process data on the device itself, allowing it to bypass security concerns that can result from data being processed through the cloud. This also enables the model to be customized more heavily to the user's liking while taking up less energy than conventional generative AI alternatives.

Samsung Gauss is currently being used internally to aid employee productivity and will be expanded into a variety of Samsung products from January 2024.