Week 25: Key AI news and surprises
Another week filled with wild new solutions, AI models, and surprises.
Meet your (perhaps) new best friend: DOT
While we impatiently wait for OpenAI to release their new speech-to-speech version of GPT 4o, work continues elsewhere on the interface between AI and humans.
This week, a new app was launched: Dot. Imagine an AI person you talk/write to who gets to know you better and better through a full memory of everything you’ve talked about and what you’ve told about yourself.
Over time, you will get a full overview of the topics and conversations you’ve had, and DOT will proactively start nudging you with various relevant tips and good advice.
The solution is also described as a form of journal/diary about yourself and your thoughts. But mostly as something we haven’t really seen before. I have started using the app myself, but I will only experience its value after some time.
Fast Company has a fascinating article about DOT: https://www.fastcompany.com/91142350/dot-an-ai-companion-app-designed-by-an-apple-alum-launches-in-the-app-store
You can try it yourself; here is a link to the app: https://apps.apple.com/dk/app/dot-living-history/id6450016041?l=da
Sutskever starts SSI
Ilya Sutskever is a name you should know. He was the technical mastermind at OpenAI and the one who got Altman fired, even though it was short-lived. Sutskever is recognized for his research in AI and for focusing heavily on the safety aspect of rolling out better and better AI on the journey towards first Artificial General Intelligence (AGI) and later Artificial Super Intelligence (ASI).
Now he has left OpenAI to dedicate himself to the development of safe superintelligence (SSI). We don’t know who pays his bills or if there is even a product or business model in his new project, but it is definitely worth following.
Read more about SSI here: https://www.theverge.com/2024/6/19/24181870/openai-former-chief-scientist-ilya-sutskever-ssi-safe-superintelligence
Link to his preliminary website: https://ssi.inc/
If you want to get to know Ilya better (and you should) and understand his thoughts, watch this 12-minute TED talk from November 2023: https://www.youtube.com/watch?v=SEkGLj0bwAU
Anthropic launches the world’s most powerful language model to date: Claude 3.5 Sonnet.
This week’s big surprise was a new AI model from the Amazon-backed company: Anthropic. It came almost out of the blue! And it’s a wild thing they’ve launched. According to Anthropic, it outperforms OpenAI’s GPT4o, which was launched just 5 weeks ago.
Tom’s Guide decided to run an advanced test of the two LLMs this week, and Claude clearly won. You should read the article, among other things, to get an insight into how advanced prompts you can make with the models currently available for free; it’s wildly impressive.
And yes, Claude 3.5 Sonnet is available to us in Denmark in a free version, so go ahead and try it: https://www.anthropic.com/
Link to the article testing GPT 4o against Claude 3.5: https://www.tomsguide.com/ai/chatgpt-4o-vs-claude-35-sonnet-which-ai-platform-wins
New soundtracks for AI videos
In recent weeks, there has been a full focus on new video generation algorithms like Kling, Lumalabs, and most recently, Runway Gen-3 Alpha, which was actually launched just recently and is in itself a huge news story (read more here: https://runwayml.com/blog/introducing-gen-3-alpha/)
Common to all these solutions is that they can generate 1–2 minute videos but without sound. And that is not always very useful.
Now Google DeepMind has launched an algorithm that solves this problem. They have created an algorithm that can analyze video and create a relevant soundtrack for it. The solution is called Deepmind V2A.
Google also has its own video algorithm called Veo (https://deepmind.google/technologies/veo/), so they know the problem internally.
You can see some wild examples of AI-generated video soundtracks here: https://deepmind.google/discover/blog/generating-audio-for-video/
IMF publishes new report on AI’s potential impact on the job market
The International Monetary Fund (IMF) released a comprehensive report this week outlining the enormous potential generative AI has for boosting productivity and changing the delivery of public services.
But they also expressed “deep concern” about the technology’s ability to disrupt labor markets and exacerbate income and wealth inequalities to an unprecedented extent.
If you don’t want to read the entire report, Perplexity has created this summary: Here are five key conclusions and findings from the IMF report “Broadening the Gains from Generative AI: The Role of Fiscal Policies”:
- The Potential and Challenges of Generative AI: Generative AI (gen AI) can significantly increase productivity and improve public services. But it also poses risks such as major labor market disruptions and increased inequality as the technology quickly automates both routine and cognitive tasks.
- The Role of Social Protection Systems: Social protection systems such as unemployment insurance and social assistance programs have historically mitigated the negative effects of automation on the labor market. These systems need to be upgraded to handle the broader and potentially more severe consequences of gen AI, focusing on expanding coverage, increasing benefit generosity, and integrating active labor market policies.
- Adjusting the Tax System: Current tax systems often incentivize investments that replace labor, such as software and computer hardware. These incentives should be reconsidered to avoid exacerbating labor market disruptions. Additionally, the taxation of capital income should be strengthened to address rising inequality and prevent erosion of the tax base as the share of labor income declines.
- Financing AI Innovation and Implementation: Public financing should focus on areas with high social returns that are unlikely to attract private investment, such as basic research, necessary infrastructure, and applications in the public sector. This is especially important for developing countries facing significant barriers to AI adoption.
- A Flexible Policy Approach: Due to the uncertainty around gen AI’s impact and speed, governments should adopt a flexible policy approach. This involves preparing for both normal and highly disruptive scenarios, ensuring that fiscal policies are flexible enough to adapt to various potential outcomes, and effectively mitigating transition costs for workers and households.
Here is a link to their blog post about the report: https://www.imf.org/en/Blogs/Articles/2024/06/17/fiscal-policy-can-help-broaden-the-gains-of-ai-to-humanity
And here is a link to the full report: https://www.imf.org/en/Publications/Staff-Discussion-Notes/Issues/2024/06/11/Broadening-the-Gains-from-Generative-AI-The-Role-of-Fiscal-Policies-549639
Mira Murati: Creative jobs will disappear
Last week, OpenAI’s CTO Mira Murati was interviewed at Dartmouth Engineering School. She talked about the upheavals we will see in almost all creative professions.
The interview is posted on YouTube and is worth spending 45 minutes on. Here is a direct link: https://www.youtube.com/watch?v=yUoj9B8OpR8