TLDR:
Runway's text-to-video model Gen-3 Alpha has been made publicly available, while similar models from OpenAI and Google remain restricted. Meanwhile, Anthropic's Claude 3.5 Sonnet language model is already outperforming industry rivals in various benchmarks. The AI job market is experiencing significant growth, with Australia projecting 200,000 AI-adjacent jobs by 2030.
In response to the increasing influence of AI in political advertising, Google has implemented new policies for disclosing AI-generated content in political ads. However, AI-powered search engine Perplexity has faced criticism for citing unreliable AI-generated sources and improperly attributing journalistic work. Microsoft's VALL-E 2 AI speech generator has achieved "human parity" but won't be released to the public due to potential ethical concerns. In a surprising turn of events, both Microsoft and Apple have withdrawn from their positions on OpenAI's board, highlighting the dynamic nature of corporate relationships in the AI industry.
This AI news update comes to you from Harriet with help from Claude 3.5 Sonnet
Runway's Gen-3 Alpha: A New Era in AI Video Creation July News
As of 1 July 2024, Runway has made its cutting-edge Gen-3 Alpha model publicly accessible, marking a significant milestone in AI-driven video generation because Sora from Open AI and Veo from Google haven't been released to the public yet. This advanced model is now available to anyone with a paid subscription, starting at $15 per month. July News
We signed up and produced the video below in less than 10 minutes!
We wanted to create a video of Janey and I in our office, so the prompt we used was:
Overhead wide-angle with dynamic motion: two women working opposite each other in an AI startup office sharing a monitor experimenting with the latest AI text to video generators, one woman is brunette and in her mid 20s, the other women is blonde and early 40s, The light is bright, and colours should be blues and greens, text should be in Neon and say 'Intelligence Assist' which is the name of the business.
The hardest part was writing the prompt using Runway's guide. The end-to-end process took about five minutes, and the results weren't bad. What do you think?
Claims from Runway about Gen-3 Alpha | Our experience |
The model produces detailed and consistent videos with realistic movements and stable elements throughout the footage. | Although the video was detailed, some details were wrong such as hands and words. |
Users can specify intricate details in their prompts, including visual descriptions, camera movements, and transitions. | You can have intricate details, but only up to 500 characters. You can vary the camera angles and movement types too. |
Gen-3 Alpha can generate videos from text, image, or video prompts. | Gen-3 Alpha image to video isn't actually available yet! Gen-2 Alpha does image to video and Gen-1 Alpha does video to video. |
The model caters to both beginners and professionals, simplifying the video creation process without requiring extensive technical knowledge. | Agreed. I have a beginner's level of video creation and found the process very easy. |
My Verdict
Playing with Runway's Gen3-Alpha was fun, and for relevant use cases like social media, virtual reality, gaming and education, I would use it more. However, for our work we felt like it wasn't applicable, so we are going to cancel our subscription. You also only get ~100 seconds of video for the credits included in the $15 subscription. Now that is very good value for money when you compare with it traditional video production but to get something usable, you'd probably burn through that quickly learning how to get the right prompt and settings.
Anthropic's Claude 3.5 Sonnet
Anthropic, a leading AI research company, has recently released Claude 3.5 Sonnet, a language model that has climbed the LMSYS leaderboard, outperforming its industry rivals in various benchmarks. Claude 3.5 Sonnet has demonstrated superior performance across a wide range of tasks, including natural language processing, question answering, and complex reasoning.
A key contributor to Claude 3.5 Sonnet's success is the prompt engineering. By analysing brief task descriptions, Claude 3.5 Sonnet can craft elaborate, nuanced prompts that yield better outcomes. This capability is important for those seeking to harness AI's full potential, as it eliminates the need for extensive manual prompt tweaking or the engagement of specialised prompt engineering experts. Essentially, Claude 3.5 Sonnet is democratising access to high-quality AI interactions, making sophisticated AI utilisation more accessible to a broader range of users.
This breakthrough highlights the rapid pace of innovation in AI technology and suggests that we may be entering a new era of more capable and versatile AI assistants. Here at Intelligence Assist, we love Claude 3.5 Sonnet and currently it’s our chosen AI model – in fact in Perplexity we’ve got it preselected to always be Claude 3.5 Sonnet.
The Growing AI Workforce: Australia's Perspective
While groundbreaking AI models like Gen-3 Alpha and Claude 3.5 Sonnet are transforming the way we work, they're also driving significant job growth in the AI sector. Recent reports from the Tech Council of Australia (TCA) provide insight into this trend:
The TCA projects that AI could lead to 200,000 AI-adjacent jobs by 2030.
Australia's AI workforce has grown from 800 workers in 2014 to over 33,000 in 2023.
To meet the 2030 goal, the AI workforce needs to expand by 500% in the next six years.
However, this rapid growth presents challenges. The TCA warns of a potential skill shortage unless significant reforms are implemented. Interestingly, AI adoption among Australian SMEs remains low, with only 19% currently utilising AI technologies as of December 2023.
AI in Political Advertising: Google's New Disclosure Policy
As AI-generated content becomes more prevalent, Google has recently announced a significant update to its political ad policies, making it easier for advertisers to disclose AI-generated content.
Under the new policy, Google will automatically generate disclosures whenever advertisers label election ads as containing "synthetic or digitally altered content." These disclosures will be visible in feeds and YouTube Shorts on mobile devices, as well as in-stream ads shown on phones, computers, TVs, and the web. For all other ad formats, advertisers will be required to include their own disclosures.
This is something the team here at Intelligence Assist will always advocate for. You may have noticed we always give credit to AI models when we've used them. By doing this, we are taking steps towards transparency, by acknowledging the potential impact of AI-generated content on public discourse and democratic processes. As AI tools like Runway's Gen-3 Alpha become more accessible and sophisticated, such policies will likely become increasingly crucial in maintaining trust and authenticity in digital media landscapes.
Emerging Challenges in AI-Powered Search with Information Reliability
While AI technology continues to advance rapidly, it also presents new challenges, particularly in the realm of information quality and reliability.
A study conducted by GPTZero, an AI content detection platform, revealed that Perplexity's search engine is drawing information from and citing AI-generated posts. So yes, AI was playing its own game of Chinese Whispers, with itself.
The study used GPTZero's AI detection software, which boasts a 97% accuracy rate, to determine if a source was AI-generated.
GPTZero CEO Edward Tian summed up the problem perfectly: "Perplexity is only as good as its sources. If the sources are AI hallucinations, then the output is too."
Microsoft's VALL-E 2 is Too Convincing for Public Release
Microsoft has developed a new artificial intelligence speech generator called VALL-E 2, which is reportedly so convincing that the company has decided not to release it to the public. This text-to-speech (TTS) generator can reproduce a human speaker's voice using just a few seconds of audio input.
According to Microsoft researchers, VALL-E 2 is capable of generating "accurate, natural speech in the exact voice of the original speaker, comparable to human performance." The AI's ability to create speech that is indistinguishable from a real person's voice has raised concerns about potential misuse, particularly in the context of voice cloning and deepfake technology. Open AI and other similar companies have also placed restrictions on their voice tech reflecting a growing awareness of the ethical implications of highly advanced AI systems.
Has there been a fallout in the AI playground? Probably not, but, here's something for my conspiracist's
In a surprising turn of events, Microsoft has dropped its seat as an observer on the board of OpenAI, less than eight months after securing the non-voting position. This move is particularly noteworthy given that Microsoft had taken the nonvoting board seat back in November, in an attempt to address questions about its interest in the startup. Adding to the intrigue, Apple, which was reportedly planning to join OpenAI's nonprofit board, will no longer do so according to reports from the Financial Times.
To put this in context:
Microsoft's board seat was a response to a turbulent period in late 2023, during which OpenAI CEO Sam Altman was fired and then promptly rehired.
These governance changes are happening against the backdrop of increasing antitrust scrutiny of the relationship between Microsoft and OpenAI.
The rapid shifts in board composition and structure highlight the volatile nature of the AI industry and its governance.
However, it's crucial to note that while these developments make for interesting industry gossip, they shouldn't significantly impact users' decisions when choosing AI tools and models. The underlying technologies and capabilities of OpenAI's products remain unchanged by these corporate governance shifts.
Conclusion
As we've seen from the latest developments - from Runway's Gen-3 Alpha in video generation to Anthropic's Claude 3.5 Sonnet in language processing, and the challenges faced by platforms like Perplexity - the AI landscape is evolving rapidly, bringing both exciting possibilities and significant challenges. It's clear that as we continue to advance AI technology, we must also prioritise the development of safeguards and quality control measures to ensure that these powerful tools benefit society as a whole.
At Intelligence Assist, we're here to help navigate all the different types of tools and models for your business, as well as how to use AI ethically and responsible.
Comments