6 things you need to know about generative AI
The World Media Group hosted its latest Smart Briefing yesterday with a panel of AI experts distilling the reality from the hype. Chaired by Economist Impact’s Emma Winchurch-Beale, the panel included Jeremy Kahn, AI Editor at Fortune, Elena Corchero, Director of Emerging Tech and Innovation at Dow Jones Live, and Hugh Langley, Senior Correspondent at Business Insider. Below is the discussion video and our key take outs from the event.
1. Cool demos – but not worth paying for yet
Insider’s Hugh Langley said it felt like there had been 500 hype cycles since the arrival of Chat GPT 18 months ago, and that would continue as the models got better. However, it’s still not clear what generative AI’s consumer proposition will be for Large Language Models (LLMs). Langley paraphrased Sam Altman, founder of OpenAI’s description of Chat GPT as a ‘cool demo, bad product’. “I think what we’re going to see with these tools is that they’re cool to play with, but not yet necessarily turning into things people are going to pay money for,” Langley says.
Langley believes we’ll start to see a more specific focus on use cases such as medicine for LLMs in the future. He said we’re also likely to see an increase in AI Agents automating processes and taking away the leg work for companies. An Agent could be used to track invoices, for example, with a minimal amount of human supervision, or an AI Travel Agent could save time building an itinerary.
2. Chat GPT still the leading model for businesses
Fortune’s Jeremy Kahn referenced a recent survey from the venture capital firm A16z, that asked 2,000 businesses about their adoption of generative AI and the models they were using. “It turned out that lots of people are playing around with lots of different models and different methodologies. But very few people actually have anything in full production,” he said.
Anything that is in full production is almost exclusively using Open AI’s GPT4 API. That may be because people are most comfortable with it as the first to market, but Kahn says it’s also still the most capable model available. “That reliability is important in enterprise use cases where being 86% accurate as opposed to 79% accurate actually makes a difference in deployment,” he says.
The vast majority of models currently in production are in employee-facing applications, not in customer or consumer-facing applications, which Kahn puts down to two things. Firstly, there’s the question of whether it’s reliable enough to be customer-facing. Secondly, the cost versus the return on investment. “These models are still very expensive to use,” he says. “People haven’t quite figured out how this is really going to be worth it.”
3. Big tech companies are winning out over start-ups
The cost is one of the problems facing start-ups. Those in the Open Source world are giving away their model for free in the hope that people will build on it. But just because you get the model for free, it doesn’t necessarily mean it’s going to save you money. ”It costs a start-up a tremendous amount to build an API and it’s not clear what their business model is so the large tech companies seem to be the big winners here,” Kahn says. Rather than investing in building AI themselves, he is noticing many organisations are waiting to see what companies like Microsoft or Salesforce create.
Despite some narratives that suggest that some of the big tech companies like Google are behind the curve on generative AI, Langley thinks they will catch up fast because they have the scale, infrastructure and brainpower to win in this space.
4. Think of AI as a layer you can add to anything
The clients that Elena Corchero works with at Dow Jones Live are enthusiastic about the potential for AI and have embraced it in the same way they have other technologies such as the Metaverse, NFT, AR and VR. Rather than it being simply the ‘next shiny thing’ however, Elena sees generative AI as the key to propelling these other technologies to the next level.
“It’s a layer that you can add to absolutely anything, so all the investment in other technologies had not been in vain,” she says, as illustrated by the Dow Jones project Sustainable Horizons. This new method of immersive storytelling demonstrates how AI can create a more sustainable world. The experience is developed using generative AI with human supervision for the script, the audio, the music, the imagery, the animation and the environments within the 3D world. It’s a brilliant example of how all the technologies come together under the layer of AI.
Kahn talked about how AI is being used in this way with self-driving cars. By layering LLM testing on top of a foundation model, the AI can explain what it’s doing while it’s driving. “It can talk through the logic of what it’s trying to do – ‘I see a pedestrian here, therefore, we’re going to apply the brakes,’” Kahn explained. This acts as a debugging technology for engineers building the systems.
It also works for financial analysis tools. Kahn cited an example of a CEO asking how to save 10 million on a budget in the next quarter. Seconds later, the AI had generated suggestions of the areas it was possible to make cuts in.
5. It’s only as good as the human controlling it
Although all our panel agreed that generative AI technology is world-changing, we shouldn’t worry about it stealing our jobs (yet). Chatbots are only as good as the search prompts you give them and much of what AI can do is flawed without human intervention. That said, employers have a responsibility to consider how they support the workforce and help people to navigate this uncertainty.
“The reality is that generative AI is going to empower people; we need to upskill them on how to use these APIs and to understand where it can fail and the flaws to look out for,” Corchero says. “We need to focus on enhancing all capabilities, but also safeguarding collective values, which is the human alignment that is missing from generative AI.”
6. Heightened vigilance and rigorous fact-checking required
The final topic centred around how to combat disinformation and misinformation, particularly in an election year. The rise of unintentional misinformation stemming from the proliferation of AI-generated content, combined with intentional disinformation, such as deep fakes, is presenting serious challenges to newsrooms.
Although some standards and legislation are in place, there is still a long way to go. “While the industry seems to be coalescing around the C2PA standard, it’s not perfect and can be manipulated,” Kahn says. Likewise, watermarks can be removed and audio content can be difficult to authenticate.
As journalists grapple with the task of distinguishing between genuine stories and fabricated or manipulated content, our panel agreed that it requires heightened vigilance and rigorous fact-checking procedures for newsrooms.
However, Corchero says this also presents an opportunity “for trusted media brands to distinguish themselves as impartial sources of truth, and to regain public trust. Now more than ever, it’s going to be very important to support people by providing that.”
Find out more about the World Media Group’s events programme or sign up to our next Smart Briefing here.