GPT-4 to launch this week, Microsoft Germany's CTO lets slip
Plus: DuckDuckGo launches its own AI web search chatbot, and more
In-brief GPT-4, the long-awaited successor to OpenAI's generative models will be unveiled next week, according to Microsoft.
Andreas Braun, CTO of Microsoft Germany, let slip that the new system will be launched next week when he was speaking at the AI in Focus - Digital Kickoff event last week.
"We will introduce GPT-4 next week, there we will have multi-modal models that will offer completely different possibilities – for example videos," he said, according to Heise, a German news site.
If the rumor is true, OpenAI's latest technology will be different from its previous models. Multi-modal systems can process different forms of data, including text, images, video, or audio. The recent ChatGPT as well as the company's older GPT-3 models can only deal with text.
Microsoft has inked contracts for exclusive access to OpenAI's models, and is steadily incorporating them into its own products. It has not revealed what model is powering its latest Bing AI web search engine yet; many people have speculated that it could be GPT-4.
DuckDuckGo is also rolling out an AI internet search chatbot
DuckDuckGo launched its own AI web search chatbot designed to answer user queries by summarizing information from Wikipedia.
Users can type a question and DuckAssist will reply and attempt to generate the answer by extracting information from the relevant Wikipedia or Britannica webpage. DuckAssist is less flexible than rivals like Microsoft's Bing. It is designed to answer specific queries and doesn't perform as well on open-ended, subjective questions.
DuckAssist is powered by text-generation models developed by OpenAI and Anthropic. "This is the first in a series of generative AI-assisted features we hope to roll out in the coming months," DuckDuckGo said in a statement. "We wanted DuckAssist to be the first because we think it can immediately help users find answers to what they are looking for faster. And, if this DuckAssist trial goes well, we will roll it out to all DuckDuckGo search users in the coming weeks.'
DuckAssist is powered by large language models, which do not always generate text that is factually correct. Wikipedia also sometimes includes false information. DuckDuckGo admitted that DuckAssist won't generate accurate answers all the time, especially if given a particularly complicated question.
"Because there's a limit to the amount of information the feature can summarize, we use the specific sentences in Wikipedia we think are the most relevant; inaccuracies can happen if our relevancy function is off, unintentionally omitting key sentences, or if there's an underlying error in the source material given."
Microsoft was the first company to launch its new AI web search chatbot for its internet browser Bing. Google is expected to deploy its rival product Bard for Google Search soon.
- Cause for a LLaMA? Meta reckons its smaller text-emitting AI is better than rivals
- Elon Musk yearns for AI devs to build 'anti-woke' rival ChatGPT bot
- Microsoft to cap daily Bing AI queries to stop the bot delivering daft responses
- Satya Nadella wants to make Google dance in battle for AI chat-powered web search
Stability AI snaps up its first company
Stability AI, a buzzy startup building text-to-image generative models, announced it was acquiring Init ML, another startup best known for developing AI image editing tools.
Init ML developed Clipdrop, software that uses machine learning algorithms to help people to easily edit and augment images. Users can, for example, scrub backgrounds, objects, and tough up the lighting, or rescale photographs using Clipdrop. These types of tools will no doubt be useful as Stability AI looks to target the creative industry with its models.
"We are excited to integrate the latest upcoming Stability AI generative models into the Clipdrop platform, enabling our technology to be easily accessible by creators everywhere," Stability AI's founder and CEO, Emad Mostaque, said in a statement.
"This acquisition will provide Clipdrop with the resources it needs to advance the next era of generative AI, marking a new milestone in our shared journey to democratize the power of AI and multimodal foundation models in creative workflows."
Headquartered in Paris, France, all of Init ML's employees will continue to work independently as a subsidiary of Stability AI. Clipdrop reportedly has over 15 million users. The financial details of the acquisition were not disclosed.
Meta to face trial over alleged trade secrets theft
Judges denied Meta's request to dismiss a lawsuit filed by Neural Magic, an AI startup, who accused the social media biz of stealing proprietary software designed to speed up machine learning models running on custom hardware.
Neural Magic claimed Meta unlawfully used its code when its first employee, Aleksandar Zlateski, left the startup to join Meta and gave his new employers knowledge of the company's algorithms. Meta, however, attempted to quash the case by arguing that Neural Magic has failed to highlight any trade secrets that can be protected, and denied Zlateski had stolen any information.
Judges at the District Court of Massachusetts, however, have ruled Meta should face trial even though all but one of the 41 secrets Neural Magic accused Meta of stealing were dismissed from the case, Reuters reported. Neural Magic is suing for $766 million in damages; the trial is expected to begin in September.
AI21 Labs releases its latest large language model
AI startup AI21 Labs launched Jurassic-2, a faster and more powerful text generation model compared to its predecessor Jurassic-1.
Jurassic-2 supports more languages other than English, including Spanish, French, German, Portuguese, Italian and Dutch, and can perform editorial tasks, like paraphrasing, summarizing, and checking the grammar of text.
"When we released Jurassic-1, we were the first company to offer open access to LLMs without any waiting list, allowing anyone to build textual generative AI apps," Ori Goshen, Co-CEO and Co-Founder of AI21 Labs, said in a statement.
"The Jurassic-2 family of models represents the next leap forward and will enable developers and organizations to build text-based applications, faster, with state-of-the-art performance at a fraction of the cost. For those who don't need all the capabilities for Jurassic-2 but still want to integrate generative AI into their websites, we also have new task-specific APIs to make our language models easily available to businesses and their needs." ®