AI Rumours and news from week 18
Another crazy week with amazing news and rumours about AI. I’ve collected the most exciting ones here:
By Tim Frank Andersen
AI video avatars with emotions
We’ve been able to create our own AI video avatars with services like HeyGen and Synthesia for a long time, but while the results are pretty impressive, the videos have been pretty expressionless. Synthesia has now changed that. This week they launched Synthesia Express-1, which reads the context of your script and adds the relevant emotions and expressions to the person. The result is significantly better, and now we’re just waiting for us to recreate ourselves too.
The company, which recently received over DKK 600 million in funding, was, by the way, founded by Danish entrepreneur Viktor Riparbelli.
See the difference in the video here: https://www.synthesia.io/avatars
And read more here: https://www.synthesia.io/post/expressive-avatars-powered-by-synthesias-new-express1-model-are-here
Co-Pilot for Microsoft 365 now available in more languages, including Danish
When Co-Pilot for Office was launched, many of us were looking forward to an easier life. But the first version unfortunately worked very poorly in Danish, so it was virtually unusable.
Now a Danish-supported version has been released, and it looks much more interesting. There’s still plenty of room for improvement, but now examples of real value are starting to emerge. It’s not cheap to get access. While the Microsoft Office package costs around £10/month, Co-Pilot costs over £2,500 for a year, which is the minimum commitment period.
But companies should probably shell out the money, because you don’t need to save many working hours per year to make it worth it.
Read more here: https://www.microsoft.com/da-dk/microsoft-365/business/copilot-for-microsoft-365?market=dk#Howtobuy
New quotes and surprises from Sam Altman
Sam Altman stopped by Stanford for a chat about AI and the future. Three days ago, Stanford posted the full interview online. As always, the OpenAI CEO is a man of wild statements, and he is one of the people in the world who knows the most about where things are headed.
“ GPT4 is the dumbest model any of you will ever have to use again… by a lot”
The conversation has already been heavily quoted. Among other things, he calls GPT-4 an incredibly stupid model. He also says that whether they burn $5 billion or $50 billion, it doesn’t matter because they are developing AGI and it will create so much more value. Finally, he believes that access to computing resources can quickly become a human right.
To watch the full interview, click here:
https://www.youtube.com/watch?v=GLKoDkbS1Cg
GPT2-Chatbot — the secret LLM
The week also held a mystery when Open-AI, without telling anyone, posted a new model called GPT2-Chatbot on a test service called LMSYS chatbot arena. Developers threw themselves at it and got surprisingly good results, especially on very complicated maths problems. This got the internet buzzing about whether it was actually a precursor to Chat GPT-5. A few days later, the model was taken down again. But Altman managed to tweet his own enthusiasm.
Here are a few links to the story:
https://www.axios.com/2024/05/02/mystery-chatbot-openai-gpt2
More rumors: OpenAI launches a Google killer search engine
Google is under pressure in search at the moment — from several sides. In Q1, Google experienced its biggest drop in market share ever.
At the same time, more and more people are starting their searches in ChatGPT. Altman has said on several occasions that there must be a better way to help people find answers than the model Google offers.
Persistent rumours now suggest that they will launch search.chatgpt.com on May 9 — a few days before Google’s big event, Google I/O, where Google usually launches all their news. Google is revolutionizing its searches with SGE (Search Generative Experience), but this is problematic because it removes a large part of its revenue potential from ads. We don’t have access to SGE in Denmark yet either.
Of course, we’ll keep a close eye on developments. Until then, you can read about the story here. https://www.tomsguide.com/ai/chatgpt/chatgpt-search-tipped-to-launch-next-week-heres-why-google-should-be-worried
AI safety board launched in the US
The US Department of Homeland Security announced this week the formation of an Artificial Intelligence Safety and Security Board, consisting of 22 members from the technology industry, government, academia, and civil liberties organizations. It’s a who’s who list of the most influential people in AI!
But given how broadly AI is defined, it’s intriguing whether the group will even be able to agree on what exactly they are protecting us from. And others think it’s a bit like putting the fox in charge of the chickens. Still, it shows that AI security is taken seriously in the US.
Link to the official page:
More about the story here: https://therecord.media/dhs-artificial-intelligence-safety-security-board-announcement
Live — directly from AI
We end with another unbelievable example of an AI generated song — this time going live from a concert, with the audience clapping and singing along all generated by AI — hear it to the end: