At Meta AI's London event this Tuesday, Yann LeCun, Nick Clegg and others discussed current topics in AI.
Clegg, the previous deputy prime minister of the UK and current president of world affairs at Meta, discussed the necessity for AI to interrupt free from the “clammy hands” of Silicon Valley.
Clegg spoke in regards to the importance of creating AI tools widely and freely available and freeing them from the monopolistic rule of a couple of large technology corporations within the US.
This is in keeping with Meta AI's ethos, which goals to challenge proprietary AI research and development at Microsoft, Google, etc. To distinguish Meta itself from “large US technology corporations” can be going too far.
While Meta's Llama series of language models just isn’t completely open source (and the meaning of the term is). hotly debated), they definitely are more open as models trained by OpenAI, Google, etc.
Meta's chief AI scientist Yann LeCun, one of the vital outstanding researchers in the sphere, also strongly supports open source AI initiatives.
“It is critical to democratize technology in order that it doesn't just remain within the cash-strapped hands of a small variety of very large and well-heeled corporations in California,” Clegg said, echoing LeCun's own sentiments, who noted, “It can’t be made by a handful of corporations on the West Coast of the United States.”
Others, like NVIDIA CEO Jensen Huang and former Stability AI Emad Mostaque, have done so talked in regards to the need for countries to construct their very own sovereign AI and free the technology from central ownership.
When I spoke at an event earlier this 12 months, Huang said“(AI) codifies your culture, the intelligence of your society, your common sense, your history – you own your individual data.”
Nick Clegg downplays AI's threat to global democracy
Clegg contradicted the prevailing narrative when he identified that AI tools haven’t been used systematically to disrupt or undermine major elections in countries reminiscent of Taiwan, Pakistan, Bangladesh and Indonesia to date this 12 months.
At first glance, it seems a bizarre comment as Indonesia, Pakistan and Bangladesh have all been affected by AI-related disinformation events. Deep fake videos targeted this in Bangladesh last 12 months Discredit oppositionfor instance, by showing how they take unpopular positions on sensitive issues reminiscent of the Israel-Gaza conflict.
In Indonesia, Erwin Aksa, deputy leader of Golkar, certainly one of Indonesia's largest political parties, said posted a deep fake by former dictator Suharto, which received over 4.7 million views. It should encourage people to vote.
Into Pakistan AI avatar of former Prime Minister Imran Khan declared victory amid a chaotic vote count that is still disputed to at the present time.
“It is true that we must be vigilant and vigilant, however it is striking how little these tools have been used systematically to really undermine and disrupt elections,” Clegg noted through the Meta AI Day event.
Clearly, we cannot easily quantify the impact or harm of AI-powered election campaigns. However, AI voting tactics have already been proven effective, and evidence from various scientific disciplines shows that they’re deep fakes have an effect on human decision-makingoften with lasting effects.
Clegg argues that AI must be viewed as each a defensive and offensive tool against disinformation or, in his words, our “sword and shield” against disinformation.
Certainly not everyone seems to be convinced of this.
Lama 3 is imminent
Meta also announced its near-term plans to launch Llama 3, its successor language model. Like its predecessors, it would be free and partially open source under Meta's own license.
Clegg announced: “Within the following month, less actually, hopefully in a really short time frame, we hope to start rolling out our recent suite of next-generation Foundation models, Llama 3.”
“Later this 12 months there will probably be a lot of different models with different capabilities and different versatility launching very soon.”
Chris Cox, Meta's Chief Product Officer, explained the corporate's vision and described its intention to integrate Llama 3 into multiple Meta products.
This model can even apparently be more open in nature, with weaker or more flexible guardrails. Meta recently withheld Emu, its image generation tool, as a consequence of concerns about latency, security and value – in order that they're not completely throwing caution to the wind.
Although details of Llama 3's parameters haven’t been revealed, it is predicted to have around 140 billion parameters, surpassing its predecessor Llama 2's 70 billion.