Microsoft Artificial Intelligence Joe Biden

Weekly AI recap: Anthropic and Inflection AI launch models poised to rival GPT-4

Author

By Webb Wright, NY Reporter

March 8, 2024 | 8 min read

Plus, more troubling political deepfake developments.

Anthropic Claude 3

Anthropic released three new versions of the AI model powering its Claude chatbot this week: Opus, Sonnet and Haiku. / Adobe Sto

Anthropic releases new AI chatbot “leading the frontier of general intelligence”

Tech companies of all sizes have been locked in a race to develop and deploy new AI chatbots since OpenAI debuted ChatGPT in late 2022, an event that could very well be remembered as the beginning of a seismic shift in Silicon Valley. In December, Google released Gemini, a large language model that’s capable of engaging not only with text but also with sound, images and other forms of content; the company announced last month that Gemini had officially subsumed Bard, an AI chatbot which Google had previously launched.

On Monday, AI start-up Anthropic – which was founded by former OpenAI founders and has since received significant financial backing from Amazon – unveiled a trio of new models, collectively dubbed Claude 3. Opus, the most powerful of the three, exceeds both OpenAI’s GPT-4 and Google’s Gemini on a range of benchmarks, including “graduate level expert reasoning” and basic math, according to Anthropic's website.

“Opus, our most intelligent model, outperforms its peers on most of the common evaluation benchmarks for AI systems,” Anthropic writes. “It exhibits near-human levels of comprehension and fluency on complex tasks, leading the frontier of general intelligence.”

Claude 3 Opus reportedly specializes in data analysis and generating computer code, making it particularly useful to scientists and software engineers. It's available now for $20 per month. A more limited version, called Sonnet, is also available for free. Both Opus and Sonnet are capable of engaging with both text-based prompts and images. The third and least advanced version, Claude Haiku, is slated to be released soon.

Inflection AI also launched an impressive new model

In other chatbot news, Inflection AI – a company founded in 2022 by DeepMind cofounder Mustafa Suleyman – unveiled a new language model this week which, like Claude 3 Opus, appears to be capable of rivaling GPT-4 in at least some domains.

The new model, called Inflection-2.5, performs particularly well in mathematical reasoning and coding (among other tasks), the company wrote in a blog post published Thursday. "We achieved this milestone with incredible efficiency: Inflection-2.5 approaches GPT-4’s performance, but used only 40% of the amount of compute for training.” Inflection-2.5 is now available for all users of Pi – a chatbot which Inflection AI released in May of last year – through iOS and Android, or through the company’s desktop app or website.

Some of the leading AI companies are positioning their chatbots as having distinct “personalities,” or modes of communication. Elon Musk’s xAI, for example, has built its chatbot Grok to be occasionally wry and humorous. Inflection AI, meanwhile, is selling Pi as an affable AI sidekick; "a kind and supportive companion that's on your side," as the company described it an a blog post announcing its launch last year.

Inflection also wrote in its Thursday blog post that Pi now has around 1mn daily and 6mn monthly active users. OpenAI announced in November that ChatGPT had around 100mn weekly active users.

Deepfakes depicting Trump with Black voters go viral

A report from the BBC published on Monday found that AI-generated images which appear to show former President Donald Trump smiling and embracing Black Americans have been spreading rampantly online. The deepfake images, according to the report, have been created by conservatives using AI tools in an effort to win Black voters over to the 2024 Trump presidential campaign.

While the images appear to be real at first glance, closer inspection reveals telltale signs of their digital origins. One of the deepfakes, for example, shows Trump smiling with his arms around two Black women, with other people in the background looking towards the “camera.” One of those people is missing a finger, and two of the fingers on Trump’s left hand are fused by a kind of flesh webbing. What's apparently supposed to be text on a man’s hat and on a woman’s shirt is also an incomprehensible mess of almost-letters.

President Biden, who's expected to run against Trump for the Oval Office in November, has also been targeted by deepfakes. In January, residents of New Hampshire were targeted with an AI-generated robocall which appeared to include the President’s voice telling them not to vote in the state's primary electio. A man who works as a magician in New Orleans reportedly later confessed to creating the deepfake audio (but not distributing it), claiming he was paid to do so by a consultant working for US Representative Dean Philips, who at the time was running against Biden for the Democratic nomination.

Generative AI-powered Adobe Express mobile app launched in beta

On Thursday, tech giant Adobe announced the beta launch of Adobe Express, a mobile app centered upon the company’s Firefly generative AI model.

Now available for free for Android and iOS devices, Adobe’s new app is described by the company in a press release as “the all-in-one, AI content creation app for everyone.” It includes features like “Generative Fill” – which allows users to insert or remove elements to or from digital images via text prompts – and “Brand Kits” – which enables marketers to upload and modify branded assets that can then be shared on social media and elsewhere.

All AI-generated assets created on Adobe Express will be attached with “Content Credentials,” according to the press release, which indicate the use of AI, the creation date and other pertinent information; the company likens it to “a digital ‘nutrition label.’”

New ML-powered tech from IAS aims to help advertisers avoid “made for advertising” sites

Adtech company Integral Ad Science (IAS) debuted a new feature on Thursday intended to help advertisers steer clear of junk-filled “made for advertising,” or MFA, sites. The idea is to use machine learning algorithms to detect and avoid such sites without hurting a campaign’s performance across other, safer channels.

​“We’re proud to offer the industry’s first pressure-tested solution for detecting and avoiding MFA and ad clutter sites at scale, while also preserving the flexibility to reach audiences across a diverse range of publishers,” IAS chief executive officer Lisa Utzschneider said in a statement.

MFA sites have been growing in number in recent years, a byproduct of a programmatic ad sector that’s incentivized by clicks over long-term impact and of the introduction of generative AI tools like ChatGPT which make it much easier to create low-quality, spammy content. One report published in June of last year by the Association of National Advertisers (ANA) revealed that more than 20% of all impressions across programmatic advertising could be traced back to MFA sites.

Microsoft’s Performance Max platform made available to all advertisers

Microsoft Advertising announced earlier this week that Performance Max (PMAX) has been deployed to all global clients. Launched in beta in July of last year, PMAX is an AI-powered marketing platform designed to optimize key components of campaign performance, including ad placement and audience targeting.

“Performance Max campaigns are designed to find and convert more customers across the entire network,” Microsoft Advertising wrote in a blog post at the time of the PMAX beta launch. “We do this in real-time by pairing the right creative assets with the right targeting parameters and creating new combinations until we find the most high-performing ad for the market. You choose the strategy, input creative assets, and articulate your goals – Performance Max leverages Microsoft AI to do the rest.”

The company hinted that advertisers can expect upgrades to PMAX in the future. “We’re investing heavily in Performance Max on our platform and are excited to continually share more updates on how we’re making the campaigns even more powerful,” Neha Mohan, principal group product manager at Microsoft, stated in a company blog post published Monday.

For more on the latest happenings in AI, web3 and other cutting-edge technologies, sign up for The Emerging Tech Briefing newsletter.

Microsoft Artificial Intelligence Joe Biden

More from Microsoft

View all

Trending

Industry insights

View all
Add your own content +