Tech / Technology

ChatGPT is down. What we know about the major outage.

Posted on:

After announcing updates to ChatGPT earlier this week, OpenAI’s chatbot is down. Here’s what we know so far.
ChatGPT logo on a smartphone

ChatGPT is down for subscribers and free users.

According to OpenAI’s status monitoring site and Down Detector (which shares the same parent company as Mashable) both reported major outages. As of 6:02 a.m. PT/9:02 a.m. ET, OpenAI says it is aware of the issue and continuing to investigate it.

Yesterday afternoon, Mashable reported issues with ChatGPT’s ability to generate DALL-E 3 images, although it’s unclear if this is a related issue. At the time a spokesperson for OpenAI said all systems are full operational on their end. Earlier this week, OpenAI announced updates to ChatGPT at its very first developer conference. This included integrating DALL-E 3 with ChatGPT, updating its knowledge cutoff date to April 2023, and deploying GPT-4 Turbo to ChatGPT Plus users.

This story is developing…

Tech / Technology

AI successfully negotiated a legal contract without human help

Posted on:

UK company Luminance successfully automated the review and negotiation of a legal contract, which could be a huge time-saver for lawyers.
A robot handing a businessman a document

In a groundbreaking moment, AI successfully negotiated a contract with…AI.

There were no real stakes involved, since it was a just a live demo, but artificial intelligence company Luminance just provided a glimpse of what the future might look like for the legal industry. On Tuesday, the UK-based company’s large language model (LLM) automated a contract negotiation “without human intervention, between two opposing parties.” Luminance claims this is the first completely AI-powered contract negotiation. The model was trained on 150 million legal documents to gain legal knowledge.

Industry-specific LLMs are one of the latest evolutions from generalist models like ChatGPT. Expect to see more of these customized AI models cropping up.

On Monday, OpenAI announced custom GPTs, which allow the user to build and train a model for their own purposes without any coding experience needed. GPTs are marketed towards ChatGPT users, but also ChatGPT Enterprise customers who use a private version of ChatGPT internally.

The benefit of using a bespoke model is not only data privacy, but also focused expertise for companies looking to automate tasks. Luminance, which does this for legal contracts, is aiming to cut down on hours lawyers spend negotiating terms.

Reviewing legal documents is a time-consuming process for lawyers. LLMs like Luminance’s proprietary legal-focused model, Autopilot, are able to understand and analyze massive amounts of information in a matter of seconds.

Automating the review of routine contracts like nondisclosure agreements could save legal professionals tons of time. “By putting the day-to-day negotiations in the hands of an AI that is legally trained and understands your business, we’re freeing lawyers up to focus their creativity where it counts,” said Jaeger Glucina, Luminance chief of staff, in the announcement.

This doesn’t mean there’s zero oversight. According to CBNC’s Ryan Browne, who saw the live demo, there’s a live log highlighting changes to clauses and suggested changes, so lawyers can review the process and double-check it for errors.

Tech / Technology

ChatGPT Plus users reporting issues since the DALL-E 3 upgrade

Posted on:

After an upgrade that integrated DALL-E 3 within ChatGPT, users are reporting issues.
ChatGPT app on a smartphone

ChatGPT users are struggling to generate images since OpenAI announced new updates.

An OpenAI spokesperson said all systems are fully operational, which matches up with its status monitoring site. Yet, reported problems spiked around 2 p.m. ET Tuesday, according to Down Detector (which is also owned by Mashable’s parent company Ziff Davis) and have continued to show high outage reports.

Mashable employees experienced errors in generating images around the same time. Users on the ChatGPT subreddit also reported issues. Several users pointed out that ChatGPT wouldn’t generate images despite the new integration with DALL-E 3.

Despite successfully generating images for Saira Mueller, a freelance editor who works with Mashable, last night, ChatGPT told her this afternoon that it didn’t have the capacity to generate images.

It then backpedaled after Mueller expressed her confusion and said it could, but didn’t actually generate anything. Instead, it mused about what to include in a prompt: “To create an image of a photorealistic landscape on another planet, please provide any specific details you’d like to include, such as the type of terrain, the presence of any celestial bodies in the sky, vegetation, or color scheme preferences.”

When Mueller followed up, it showed an error message.

ChatGPT showing an error message

A Mashable editor ultimately got an error message after ChatGPT failed to generate an image.
Credit: Screenshot: Mashable / OpenAI

At OpenAI’s developer conference on Monday, CEO Sam Altman announced ChatGPT would be doing away with the model selection dropdown, which featured the DALL-E, internet browser, and plugins options. Instead these would all be integrated within the standard interface for ChatGPT Plus users.

Users with the updated integration are the ones experiencing the issues. It’s unconfirmed whether OpenAI has deployed the updates, causing the outages, but given the timeline announced by Altman yesterday, this would match up.

Mashable has reached out to OpenAI for more information and will update this story if it hears back.