The week in AI: OpenAI claims The New York Times 'hacked' ChatGPT for lawsuit

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each Thursday, we aggregate the major developments in artificial intelligence; we pass along the news, useful resources, tools and services, and highlight the top research in the field as well as exciting developments in open source. Even if you aren’t an engineer, we’ll keep you in touch with what’s going on in AI.

NEWS & OPINION

OpenAI accuses The New York Times of ‘hacking’ ChatGPT to set up lawsuit

-------------------------

The New York Times vs. OpenAI lawsuit is heating up. In a court filing on Monday, OpenAI claimed that NYT paid a ‘hacker’ to exploit ChatGPT. From the filing:

“The allegations in the Times’s Complaint do not meet its famously rigorous journalistic standards. The truth, which will come out in the course of this case, is that the Times paid someone to hack OpenAI’s products. It took them tens of thousands of attempts to generate the highly anomalous results that make up Exhibit J to the Complaint. They were able to do so only by targeting and exploiting a bug (which OpenAI has committed to addressing) by using deceptive prompts that blatantly violate OpenAI’s terms of use. And even then, they had to feed the tool portions of the very articles they sought to elicit verbatim passages of, virtually all of which already appear on multiple public websites. Normal people do not use OpenAI’s products in this way.”

OpenAI claims that The Times used deceptive prompts (i.e. repeatedly asking ChatGPT, "what's the next sentence?") - with accompanying NYT articles - to target training data regurgitation and induce model hallucination.

OpenAI has argued that the court should dismiss claims alleging direct copyright, contributory infringement, and other "legally infirm” claims. AI is a new arena for copyright disputes; the judges in these cases are going to have their work cut out for them in setting precedents. In our estimation, this argument doesn’t look like OpenAI’s best defense. It’s on them to provide the working guardrails - and how ‘normal people’ use their product seems immaterial.

Apple abandons secretive fully self-driving EV project after ten years, will ‘break new ground’ in generative AI this year

-------------------------

Apple has abruptly canceled its plans to release an electric car with self-driving abilities, a secretive product that had been in the works for nearly a decade. According to reports, the company told employees in an internal meeting on Tuesday that it had scrapped the project and that members of the group would be shifted to different roles - primarily to Apple’s generative AI divisions.

Tesla CEO Elon Musk appeared to be pleased about the report, tweeting a salute emoji with a burning cigar in response.

Separately, Apple CEO Tim Cook promised shareholders during the company’s annual meeting that Apple would “break new ground” on GenAI this year. What that ground might be is anyone’s guess at this point (Apple has been much more quiet on the AI front than its big tech rivals), but there are hints out there about some of the things we can expect. We know that Siri and iOS’ built-in search tool Spotlight are getting AI makeovers. Their LLM in development, Ajax, is expected to be on par with ChatGPT or better. Apple has also recently (finally) been more active in published research and open source AI: projects like HUGS and MGIE will likely end up transforming how we interact with visual media.

Whatever the company ends up having to offer in 2024, it’ll be great to see more Apple in the AI space.

Klarna’s AI does the work of 700 humans better than 700 humans

-------------------------

Swedish payment provider Klarna announced this week that its AI assistant, powered by OpenAI, handled two-thirds of all the company’s customer service chats over a month-long period. The AI handled 2.3 million conversations in that time, which is reportedly equivalent to the work of 700 full-time employees.

Customer satisfaction ratings for the AI service was on par with human agents, but the error rate in resolving queries actually declined by 25 percent with the AI. Additionally, Klarna customers now resolve their issues in less than two minutes, compared to eleven minutes previously.

Klarna CEO Sebastian Siemiatkowski discussed the results on X, highlighting the current (not future) issue of AI's impact on job markets. "We decided to share these statistics to raise the awareness and encourage a proactive approach to the topic of AI. For decision makers worldwide to recognise this is not just ‘in the future’, this is happening right now," Siemiatkowski wrote.

• Google hopeful of fix for Gemini’s historical-image diversity issue within weeks

• Voters are getting calls from freakishly smart AI bots. I talked to one for 10 minutes

• Social media users struggle to spot political AI bots

• Google paying independent publishers to test unreleased generative AI platform

• What happens when we outsource boring but important work to AI? Research shows we forget how to do it ourselves

• AI ‘dream girls’ are coming for porn stars’ jobs

• Ghost kitchens are advertising AI-generated food on DoorDash and Grubhub

• Nvidia CEO Jensen Huang says kids shouldn't learn to code - they should leave it up to AI

• How many news websites block AI crawlers?

• Tumblr and WordPress posts will reportedly be used for OpenAI and Midjourney training

TECHNICAL, RESEARCH & OPEN SOURCE

Mistral releases flagship model Mistral Large, announces partnership with Microsoft

-------------------------

French AI company Mistral, the creators of popular open source models Mistral 7B and Mixtral 8×7B, is looking to Microsoft to help them commercialize their newest cutting-edge model, Mistral Large. Mistral Large scored an 81.2% on the MMLU (Massive Multitask Language Understanding) benchmark, closer to GPT-4’s 86.4% than Google’s Gemini Pro or Anthropic’s Claude. The context window is 32k tokens, which is smaller than some competitors (128k for GPT-4 Turbo, 100k for Claude).

You can test the Mistral Large model for free with Mistral’s new ChatGPT-like conversational chatbot, Le Chat (message limits may apply for the Large model). The Mistral Large API is available on Mistral’s own infrastructure, hosted in Europe, or through Azure AI Studio and Azure Machine Learning. Mistral Small will also be available today, offering improved latency over Mistral’s 8x7B model. Mistral Large is priced at $0.008 per 1k tokens for inputs and $0.024 per 1k tokens for outputs, slightly cheaper (about 20%) than GPT-4 turbo.

Phind-70B: closing the code quality gap with GPT-4 Turbo while running 4x faster

-------------------------

Phind’s newest coding assistant LLM, 70B, outperforms GPT-4 Turbo in HumanEval coding benchmarks and speed. Phind-70B is based on the CodeLlama-70B model and fine-tuned on an additional 50 billion tokens, It achieves a HumanEval score of 82.3% and has a context window of 32K tokens.

You can try Phind for free without a login - their UI has both a search and a chat function; for writing/manipulating code the chat function seems to provide better results. They claim their model is less lazy than GPT-4 and doesn’t hesitate to generate detailed code examples. Phind plans to open source their 34B model soon, with the 70B model to follow at a later date. Phind is compatible in Visual Studio Code.

MORE IN T/R/OS

• EMO: Emote Portrait Alive - generating expressive portrait videos with Audio2Video diffusion model under weak conditions

• Google DeepMind’s Genie: Generative Interactive Environments

• Microsoft releases its internal generative AI red teaming tool to the public

• StarCoder2: family of open coding LLM models trained on the newest, largest open coding dataset (The Stack v2)

That’s it for this week! We’ll see you next Thursday.