French police raid Nvidia offices amid antitrust investigation: Report

September 29, 2023 Coin Telegraph

Amazon

The action reportedly came as a part of a general inquiry of French antitrust authorities into the cloud computing sector.

Nvidia, one of the global leaders in artificial intelligence (AI) chip production with headquarters in California, reportedly faced a police raid in its French offices this week. The action came as a part of a general inquiry of French antitrust authorities into the cloud computing sector.

The Wall Street Journal reported the raid on Sept. 28, although neither Nvidia nor the French enforcement agencies have officially commented on what happened.

A press release on the webpage of the French antitrust agency, Autorité de la Concurrence, refers to an unannounced inspection in the graphics cards sector. According to the release, a judge authorized the raid at the premise of the company “having implemented anticompetitive practices in the graphics cards sector.”

However, the raid itself does not “pre-suppose the existence of a breach of the law, which could be imputed to the company,” as the message from the agency specifies.

Autorité de la Concurrence refers to its own opinion, issued in conclusion to a year-long study of the cloud computing sector. Published in June 2023, this document does not mention Nvidia. Instead it focuses on other tech companies, namely the three hyper scalers” — Amazon Web Services (AWS), Google Clou,d and Microsoft Azure. According to the agency’s data, they represent 80% of the spending growth in public cloud infrastructures and applications in France in 2021:

“Amazon and Microsoft have captured 46% and 17% respectively of revenues from IaaS and PaaS services in 2021. Given their financial capacities and their digital ecosystems, these hyperscalers are in a position to hinder competition development.”

The agency is considering various options provided by the national competition laws and the European Data Act to combat this tendency.

Nvidia inevitably comes under the regulators' attention due to its unique position as the hardware producer for the most innovative sectors of the digital industry. The company's recent quarterly report revealed that the United States regulators asked it to curb exports of AI chips to “some Middle East countries.” A day later, the United States Department of Commerce denied this information.

ChatGPT can now browse the internet, no longer limited to info from 2021

September 27, 2023 Coin Telegraph

chatGPT

The browsing update is limited to Plus and Enterprise subscribers for now but will roll out to all users “soon,” says OpenAI.

ChatGPT can now browse the web and integrate up-to-date information into its responses, according to a Sept. 27 post on X (formerly Twitter) from artificial intelligence (AI) firm OpenAI.

ChatGPT can now browse the internet to provide you with current and authoritative information, complete with direct links to sources. It is no longer limited to data before September 2021. pic.twitter.com/pyj8a9HWkB
— OpenAI (@OpenAI) September 27, 2023

The updates are available immediately for Plus and Enterprise users using the GPT-4 model, according to the post. The web browsing feature will be available for other users “soon,” but OpenAI didn’t specify whether that meant GPT-4 would be enabled for non-premium users or if browsing would be implemented for the GPT 3.5 model.

OpenAI did not immediately respond to a request for clarification.

Per the post:

“ChatGPT can now browse the internet to provide you with current and authoritative information, complete with direct links to sources. It is no longer limited to data before September 2021.”

Before this update, ChatGPT suffered from an ever-widening gap in its knowledge base. Due to the nature of how AI models such as generative pre-trained transformers (GPT) are trained, ChatGPT’s knowledge base previously ended in 2021 — presumably the year it was essentially finalized for production.

In previous testing, OpenAI reported successes and setbacks in its attempts to implement browsing with ChatGPT. In June of 2023, reports began to surface that beta versions of ChatGPT with browse were being used to bypass paywalls. It also appeared to have a penchant for hallucinating and sometimes conflating recent and historical information.

ChatGPT’s browsing feature update follows directly on the heels of OpenAI’s Sept. 25 announcement that the chatbot was also receiving a multimodal update:

ChatGPT can now see, hear, and speak. Rolling out over next two weeks, Plus users will be able to have voice conversations with ChatGPT (iOS & Android) and to include images in conversations (all platforms). https://t.co/uNZjgbR5Bm pic.twitter.com/paG0hMshXb
— OpenAI (@OpenAI) September 25, 2023

As Cointelegraph reported, ChatGPT, GPT-3.5 and GPT-4, will be able to understand plain language spoken queries and respond in one of five different voices as well as view and interpret images.

OpenAI also recently announced DALL-E 3, the latest version of its image-generation AI.

With the slew of updates sweeping across the company’s products, it’ll be interesting to see what it has left to reveal at its first-ever developer’s conference, OpenAI DevDay, slated to take place on Nov. 6 in San Francisco.

CIA to build its own ChatGPT-style AI bot for investigations: Report

September 27, 2023 Coin Telegraph

chatGPT

The CIA says it is creating its own ChatGPT AI tool to conduct open-source investigations by analyzing public information.

The United States Central Intelligence Agency (CIA) is planning to build and deploy its own ChatGPT-style artificial intelligence (AI) bot for investigations, according to a report from Bloomberg.

The CIA reportedly plans to equip its analysts with the new AI tool to better access open-source intelligence. This includes sifting through public information for leads in investigations.

Agency officials said they intend to roll out the tool soon. Randy Nixon, director of the CIA’s open source enterprise, told Bloomberg that they have gone from utilizing “newspapers and radio, to newspapers and television, to newspapers and cable television, to basic internet, to big data, and it just keeps going.”

The development comes after critics have described the current methods through which the CIA processes available public data as “slow.”

Bloomberg reported that the new tools will allow its users to see the original source of the information being sourced, along with a chat feature.

“Then you can take it to the next level and start chatting and asking questions of the machines to give you answers, also sourced.”

There was no mention of the model from which the CIA is building its new tool nor its privacy protections, though Nixon said it “closely follows” privacy law in the United States.

The tool will reportedly be available across the 18 agencies that make up the U.S. intelligence. This includes the CIA, National Security Agency, Federal Bureau of Investigation and military-run agencies.

This development from the CIA comes after recent confirmation from other governmental agencies of the use of AI.

On Sept. 12, Gary Gensler, the U.S. Securities and Exchange Commission (SEC) chair, confirmed to the U.S. Senate that his department is currently employing AI tools to monitor the finance industry to search for clues of fraud and manipulation.

In a speech on July 17, Gensler praised the use of AI tools, saying SEC staff could benefit from greater use of AI in “market surveillance, disclosure review, exams, enforcement and economic analysis.”

Magazine: ‘AI has killed the industry’: EasyTranslate boss on adapting to change

ChatGPT can now speak, listen and see images

September 25, 2023 Coin Telegraph

chatGPT

OpenAI collaborated with professional voice actors to train the models to speak.

The generative artificial intelligence (AI) space continues to heat up as OpenAI has unveiled GPT-4V, a vision-capable model, and multimodal conversational modes for its ChatGPT system.

With the new upgrades, announced on Sept. 25, ChatGPT users will be able to engage the chatbot in conversations. The models powering ChatGPT, GPT-3.5 and GPT-4, can now understand plain language spoken queries and respond in one of five different voices.

ChatGPT can now see, hear, and speak. Rolling out over next two weeks, Plus users will be able to have voice conversations with ChatGPT (iOS & Android) and to include images in conversations (all platforms). https://t.co/uNZjgbR5Bm pic.twitter.com/paG0hMshXb
— OpenAI (@OpenAI) September 25, 2023

According to a blog post from OpenAI, this new multimodal interface will allow users to interact with ChatGPT in novel ways:

“Snap a picture of a landmark while traveling and have a live conversation about what’s interesting about it. When you’re home, snap pictures of your fridge and pantry to figure out what’s for dinner (and ask follow up questions for a step by step recipe). After dinner, help your child with a math problem by taking a photo, circling the problem set, and having it share hints with both of you.”

The upgraded version of ChatGPT will roll out to Plus and Enterprise users on mobile platforms in the next two weeks, with follow-on access for developers and other users “soon after.”

ChatGPT’s multimodal upgrade comes fresh on the heels of the launch of DALL-E 3, OpenAI’s most advanced image generation system.

According to OpenAI, DALL-E 3 also integrates natural language processing. This allows users to talk to the model in order to fine-tune results and to integrate ChatGPT for help in creating image prompts.

In other AI news, OpenAI competitor Anthropic announced a partnership with Amazon on Sept. 25. As Cointelegraph reported, Amazon will invest up to $4 billion to include cloud services and hardware access. In return, Anthropic said it will provide enhanced support for Amazon’s Bedrock foundational AI model along with “secure model customization and fine-tuning for businesses.”

Amazon invests $4 billion Anthropic AI startup

September 25, 2023 Coin Telegraph

Amazon

Amazon and the AI startup Anthropic have entered into an investment agreement that invests $4 billion into the start-up to develop high-performing foundation models.

Amazon and artificial intelligence (AI) startup Anthropic have announced a new investment agreement to aid the research and development of new high-performing foundation models.

In a post on X, formerly Twitter, Anthropic revealed that Amazon has invested $4 billion in its work, while receiving access to Amazon cloud services (AWS) Trainium and Inferentia chips.

It said it will be offering in return “enhanced support” of Amazon Bedrock, which produces foundation models, with "secure model customization and fine-tuning" for businesses. Amazon teams will also be able to use Bedrock to build on Anthropic’s models.

Our agreement provides Anthropic with access to @awscloud Trainium and Inferentia chips. We’ll also offer enhanced support of Amazon Bedrock with secure model customization and fine-tuning for businesses.
— Anthropic (@AnthropicAI) September 25, 2023

Additionally, through the deal Amazon is reported to be taking a “minority stake” in Anthropic. The latter said this has caused no disruption in its governance.

“As outlined in this policy, we will conduct pre-deployment tests of new models to help us manage the risks of increasingly capable AI systems.”

The AI startup was formed by former members of the Microsoft-backed OpenAI, the creator of the viral AI chatbot ChatGPT.

This latest development comes shortly after Anthropic announced an investment of $100 million from the South Korean telecommunications giant SK Telecom back in August.

That investment followed a collaboration between Anthropic and SK Telecom to develop a multilingual large language model (LLM) for the latter’s Telco AI Platform.

Anthropic has also recently been a part of major movements within the AI community. In July it joined Google, OpenAI, Microsoft and others in the formation of the “Frontier Model Forum,” which was created in order to self-regulate the development of from the inside.

It has also been a part of initiatives led by the United States government relating to AI development and regulation, including a cybersecurity challenge to help strengthen its “critical infrastructure.”

Magazine: ‘AI has killed the industry’: EasyTranslate boss on adapting to change

The Author’s Guild launches class-action lawsuit against OpenAI

September 20, 2023 Coin Telegraph

Coin Telegraph

The Author’s Guild opened a lawsuit against OpenAI, alleging misuse of copyrighted material in training of its AI models.

The Author’s Guild in the United States opened a class-action lawsuit against the Microsoft-backed OpenAI on Sept. 19 due to its alleged misuse of copyrighted material in the training of its artificial intelligence (AI) models.

According to court documents, the oldest and largest professional organization for writers in the U.S. is operating under the Copyright Act and seeking “redress” for what it calls “flagrant and harmful infringement” of registered copyrights in written works of fiction.

It goes on to argue that works were copied wholesale and without permission or “consideration” by feeding them into large language models (LLMs).

“These algorithms are at the heart of Defendants’ massive commercial enterprise. And at the heart of these algorithms is systematic theft on a mass scale.”

The Author’s Guild said it represents a class of professional fiction writers whose “works spring from their own minds and their creative literary expression.” It says, therefore, that since their livelihoods derive from these creative works, the LLMs “endanger” the ability of fiction writers to make a living.

It suggested that the AI models could’ve been trained via the public domain, or OpenAI could have paid a licensing fee for the usage of the copyrighted works.

“What Defendants could not do was evade the Copyright Act altogether to power their lucrative commercial endeavor, taking whatever datasets of relatively recent books they could get their hands on without authorization.”

On Sept. 11, the Guild posted an article on X about how authors can protect their work from AI web crawlers.

Did you know that you can block OpenAI's web crawler from accessing their website? We've gathered some practical tips for authors to utilize so that they may protect their work from AI. https://t.co/6PwAre6hFt
— The Authors Guild (@AuthorsGuild) September 11, 2023

Pinned to the top of its profile, the Author’s Guild has a link to its advocacy work in regards to AI technologies.

This filing from the Author’s Guild follows updates in a similar lawsuit against Meta and OpenAI and their respective AI models using copyrighted material in training.

Author Sarah Silverman and others opened the lawsuit in July; however, now both companies have asked judges to dismiss the claims.

In August, the U.S. Copyright Office issued a notice of inquiry on AI, seeking public comment on topics related to AI content production and how it should be handled by policymakers when AI content mimics that which is made by human creators.

Prior to the inquiry, U.S. District Judge Beryl Howell ruled that artwork created solely by AI is not eligible for copyright protection.

Magazine: NFT collapse and monster egos feature in new Murakami exhibition

Meta refutes claims of copyright infringement in AI training

September 20, 2023 Coin Telegraph

Coin Telegraph

In a lawsuit against Sarah Silverman and other authors Meta claims its AI system does not create copyright infringing material.

Meta has refuted claims that its artificial intelligence (AI) model Llama was trained using copyrighted material from popular books.

In court on Sept. 18 Meta asked a San Francisco federal judge to dismiss claims made by author Sarah Silverman and a host of other authors who have said it violated copyrights of their books in order to train its AI system.

The Facebook and Instagram parent company called the use of materials to train its systems “transformative” and of “fair use.”

“Use of texts to train LLaMA to statistically model language and generate original expression is transformative by nature and quintessential fair use..."

It continued by pointing out a conclusion in another related court battle, “much like Google’s wholesale copying of books to create an internet search tool was found to be fair use in Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015).”

Meta said the “core issue” of copyright fair use should be taken up again on, “another day, on a more fulsome record.” The company said the plaintiff couldn’t provide explanations of the “information” they’re referring to, nor could they provide specific outputs related to their material.

The attorneys of the authors said in a separate statement on Sept. 19 that they are "confident” their claims will be held and will continue to proceed through “discovery and trial.”

OpenAI also attempted to dismiss parts of the claims back in August under similiar grounds to what Meta is currently proposing.

The original lawsuit against Meta and OpenAI was opened in July and was one of many lawsuits popping up against Big Tech giants over copyright and data infringement with the rise of AI.

On Sept. 5 a pair of unnamed engineers opened a class-action lawsuit against OpenAI and Microsoft regarding their alleged scraping methods to obtain private data while training their respective AI models.

In July, Google was sued on similar grounds after it updated its privacy policy. The lawsuit accused the company of misusing large amounts of data, including copyrighted material, in its own AI training.

Magazine: AI Eye: Real uses for AI in crypto, Google’s GPT-4 rival, AI edge for bad employees

Alibaba launches its ChatGPT-like AI model for public use amid loosening restrictions in China

September 13, 2023 Coin Telegraph

Alibaba

The launch comes just weeks after Chinese authorities began accepting applications for public-facing AI systems.

Alibaba announced that its proprietary large language model, an artificial intelligence system called Tongyi Qianwen, will be available for public and enterprise access throughout China starting Sept. 13.

Tongyi Qianwen is a ChatGPT-like large language model trained on a corpus of English and Chinese text. While its exact specifications aren’t known — early rumors indicated it would be trained with as many as 10 trillion parameters, 10 times as many as OpenAI’s GPT4, but these remain unsubstantiated — Alibaba previously released two 7 billion-parameter open-source models based on the Tongyi Qianwen architecture.

Alibaba Cloud launched two open-source large vision language models, Qwen-VL and Qwen-VL-Chat, that can comprehend images, texts, and bounding boxes in prompts and facilitate multi-round question answering in both English and Chinese.

Learn more: https://t.co/LrVNzNIRVy… pic.twitter.com/0t0Et6j1BU
— Alibaba Cloud Intelligence Group (@cloudintelgrp) August 30, 2023

Previously, Tongyi Qianwen had only been available to a limited group of users during its beta test phase. The public rollout coincides with a recent loosening of restrictions related to the use of artificial intelligence (AI) technologies in the People’s Republic of China.

A set of guidelines published by the Chinese government in June dictated that, going forward, all AI technologies released to the public would require a special vetting and certification process.

The rules went into effect on Aug. 15. As Cointelegraph reported, a number of Chinese companies were given approval to launch models, including Baidu, Tencent, TikTok and ByteDance.

Among the provisions included in the updated restrictions are rules barring the generation of images in the likeness of China’s president, Xi Jinping, and mandates indicating organizations will address objectionable content within a three-month period. Previous versions of the legislation called for monetary fines, but as Cointelegraph reported, those plans were axed.

As China explores a loosening of its regulations, the United States has taken only preliminary steps to regulate AI technologies. Most recently, on Sept. 13, Senate Minority Leader Chuck Schumer hosted a litany of top U.S. tech CEOs and founders in the first of nine scheduled forums to discuss potential policy ideas.

Meta is building AI model to rival OpenAI’s most powerful system

September 11, 2023 Coin Telegraph

chatGPT

Meta is reportedly in the process of building a new, more powerful and open-source AI model to rival the most powerful systems of its rival OpenAI.

Meta, the parent company to social media platforms Facebook and Instagram, says it's developing a new artificial intelligence (AI) that will rival the most advanced model from OpenAI, according to a Wall Street Journal exclusive.

WSJ reported that individuals familiar with the matter said Meta aims for the new AI model to be “several times” more powerful than its Llama 2 model, which it released earlier this year.

For the moment, the WSJ sources say Meta’s plans for the new system are for it to be open-source, and therefore allow other companies to build AI tools to produce high-level text, analysis and other types of output.

The company has also been building data centers necessary to create such a high-level system while acquiring more of Nvidia’s H100 semiconductor chips - the most powerful and coveted chips currently available on the market.

Llama was trained on 70 billion parameters, and while OpenAI hasn’t released its parameters for GPT-4; it's estimated around 1.5 trillion.

The sources said Meta anticipates training to begin for the large language model (LLM) in early 2024 and to be ready for release sometime next year. It is likely to be released after Google’s expected forthcoming LLM Gemini.

Microsoft is a primary backer of OpenAI and also collaborated with Meta to help make Llama 2 available on Azure, its cloud-computing platform. However, the sources said Meta plans to train its upcoming model on its own infrastructure.

This development comes as major tech companies and governments are racing to create, deploy and control high-level AI systems.

Recently, the United Kingdom government announced that it plans to spend $130 million on high-powered chips to create AI systems.

Across the globe in China, the country’s new legislation on AI recently went into effect. Since then the CEO of Baidu, a major China-based tech company, said that over 70 AI models have been released in the country.

Magazine: AI Eye: Get better results being nice to ChatGPT, AI fake child porn debate, Amazon’s AI reviews

ChatGPT web traffic drops for third consecutive month in August

September 8, 2023 Coin Telegraph

Coin Telegraph

Web traffic from users worldwide for OpenAI’s ChatGPT declined during the summer months, which could be linked to university-aged users out of school.

The popular artificial intelligence (AI) chatbot ChatGPT has seen a decrease in user traffic worldwide for the third consecutive month this August, according to data from analytics company Similarweb.

In August, desktop and mobile site traffic to OpenAI’s popular chatbot from visitors worldwide decreased by 3.2% to 1.43 billion. This follows a 10% drop in traffic two months prior. In addition, the amount of time spent on the site also dropped, though slightly, from 8.7 minutes to 7 minutes in August.

However, visits stemming from United States-based users began to increase in August by 0.4%. Unique visitors to the site, which slumped in June and July, rose by 3% in the U.S. and 0.3% worldwide in August.

Similarweb senior insights manager David F. Carr, who consistently tracks AI chatbots and authored the report, wrote that the fluctuation in users could be the result of students using the program having summer break and now resuming classes.

“Students seeking homework help appears to be part of the story: the percentage of younger users of the website dropped over the summer and is now starting to bounce back.”

This theory can be backed up by the drop in audience for the summer months of ChatGPT users in the 18–24 age range, both in the U.S. and worldwide.

According to the data, in the 18–24 age bracket in the U.S., traffic dropped 10% in May, around the time U.S. universities finish their semester, 15% in June, and another 4% in July. At its peak in April, 18–24-year-old visitors made up 30% of the total audience share in the United States.

*ChatGPT users in the 18–24 age bracket in the United States. Source: Similarweb*

*ChatGPT users in the 18–24 age bracket worldwide. Source: Similarweb*

A separate survey from May 2023 by Intelligent.com surveyed 1,223 undergraduate and graduate students in the U.S. and found that 30% answered that they had used ChatGPT for schoolwork during the academic year.

It reported that of those, 46% “frequently” use the tool for homework, and 1 in 8 said they saw an increase in GPA, which could be traced to their usage of the AI chatbot.

At the moment, there are no overarching rules regarding AI usage in universities within the United States. However, in Japan, the Ministry of Education has already spoken out on its plans to allow limited use of generative AI tools in elementary, junior high and high schools.

Magazine: Tencent’s AI leviathan, $83M scam busted, China’s influencer ban: Asia Express

openai