It's been almost a week since Sundar Pichai, the CEO of Google, issued a code red over OpenAI's GPT-3, recognizing the potential impact it could have on Google's search engine business. But let me tell you this, Google's AI is far from being outdated. In fact, they have not one, not two, but three language models in the works that are set to revolutionize the world of technology. And these models are just the tip of the iceberg, as Google has a plethora of other state-of-the-art AI models for fields such as code and image generation.
Let's start with Sparrow, the chatbot that is the weakest of the three language models from Google. Sparrow is based on the Chinchilla model, which has only 70 billion parameters. However, due to Google's advanced training techniques, Sparrow is outperforming GPT-3, which has 170 billion parameters, on every benchmark test.
Aspiro is a fine-tuned version of Chinchilla, just like OpenAI's GPT-3 is a fine-tuned version of GPT-3.5. Aspiro and GPT-3 are almost on the same level, but Aspiro has a key advantage – it is trained to provide evidence for its answers and links when asked by the user. This seemingly small difference is a huge win for Aspiro, as the inability to validate an answer is holding many users back from fully utilizing AI technology.
Aspiro was created by DeepMind, a subsidiary of Alphabet, who have been at the forefront of AI research for over a decade. DeepMind is the same company that created AlphaGo, AlphaZero, and AlphaFold, which changed the entire AI research landscape with each release. In 2022, DeepMind was even recognized by Science Magazine's Top 10 Breakthroughs for their AlphaCode and AlphaTensor products – AlphaCode being the best programming AI in the world, and AlphaTensor being an AI for efficient matrix multiplication.
But Google's AI doesn't end there. There is a system called Lambda, which contains a language model named Mina. Mina is quite comparable to GPT-3 and is the language model inside Lambda. However, Lambda has many other modules, making it much more advanced than GPT-3. In fact, Google engineer Blake Lemoine, who was fired from Google for claiming that Lambda was sentient, stated in an interview on the H3 podcast that "GPT-3 is just a module inside Lambda."
Lambda's capabilities are mind-blowing. If you were to have a conversation with GPT-3 and ask it about a movie, it would simply respond with "I don't know" if the knowledge about that movie wasn't in the model. With Lambda, if you were to ask it about a movie, it can query the web, watch the movie, and learn from it. This is made possible by Google's crawlers, which act as the eyes and ears of Lambda. The entire community of SEO experts is trying their best to help Google understand the content of a page by using HTML tags, providing accurate descriptions, and optimizing for search engines. Now, it's just a matter of feeding this information to Lambda to make it even more advanced.
And this is just the tip of the iceberg. With lambda, Google has the potential to create an AI that can understand the content of a page and even its intent. This is a huge leap forward in the world of AI and search engines. And with this, Google is positioning itself to be the leader in the field.
So why should you care about all this? Well, think about it. What if you could have a search engine that could understand what you're looking for, not just based on keywords, but on the context and intent behind your search? What if you could have a search engine that could give you not just the information, but also the evidence to support it? This is the kind of search engine that Google is working on with lambda.
And it's not just limited to search engines. The potential applications of this technology are vast and limitless. Think about a chatbot that could answer any question you have with accurate and reliable information. Think about a programming AI that could write code for you that is not only correct, but also efficient and optimized. Think about a language model that could generate creative and meaningful content for you, without the need for manual input. These are all possible with the advancements that Google is making in the field of AI.
So, if you're still thinking that Google is the one in trouble in the age of AI, think again. Google is not just staying ahead of the curve, they're shaping the curve. With their state-of-the-art AI models and advancements, they're setting the bar for the rest of the industry to follow. And I, for one, can't wait to see what they come up with next.
In conclusion, Google is not just a company, it's a pioneer in the field of AI. And with lambda, they're about to change the world as we know it. So sit back, buckle up, and get ready for the ride of your life, because Google's AI is coming, and it's going to be one wild ride.
Here are some sources for the mentioned Google AI models:
Sparrow is a machine learning framework for numerical optimization problems. It is a research project from Google Brain Team. Information on Sparrow can be found on Google Research's website: https://research.google/teams/brain/
Aspiro is a research project from the Google Brain team that focuses on audio generation and manipulation. Information on Aspiro can be found on Google Research's website: https://research.google/teams/brain/
DeepMind is a leading artificial intelligence company that was acquired by Google in 2015. It is widely recognized for its breakthroughs in AI, such as AlphaGo, AlphaZero, and AlphaFold. The official website for DeepMind is: https://deepmind.com/
Alphabet is the parent company of Google and many other subsidiary companies. It was created as a holding company for Google in 2015. The official website for Alphabet is: https://abc.xyz/
AlphaGo is a computer program developed by DeepMind that is capable of playing the board game Go. It made headlines in 2016 when it defeated the world champion of Go, Lee Sedol. Information on AlphaGo can be found on the DeepMind website: https://deepmind.com/research/case-studies/alphago-the-story-so-far/
AlphaZero is a machine learning system developed by DeepMind that uses reinforcement learning to learn and play games such as chess, shogi, and Go. Information on AlphaZero can be found on the DeepMind website: https://deepmind.com/research/case-studies/alphazero-shedding-new-light-on-the-grand-games-of-chess-shogi-and-go/
AlphaFold is a protein folding prediction system developed by DeepMind. It was introduced in 2018 and won the Breakthrough Prize in Life Sciences in 2019. Information on AlphaFold can be found on the DeepMind website: https://deepmind.com/research/case-studies/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology/
AlphaCode is an AI system developed by DeepMind that is capable of searching a vast solution space through ingenious design, large amounts of data, and significant computing power. The website for AlphaCode provides the ability to hover over tokens in the solution and see the tokens the model attended to during solution generation.
https://bdtechtalks.com/2022/02/07/deepmind-alphacode-competitive-programming/AlphaTensor is an AI system from DeepMind for discovering new algorithms for tasks like matrix multiplication with reinforcement learning. It sheds light on a 50-year-old open question in mathematics and provides provably correct algorithms. https://www.deepmind.com/blog/discovering-novel-algorithms-with-alphatensor
Lambda language model Mina: Mina is a recently published research paper by Google AI on a language model called "Lambda." Information on the Lambda language model and Mina can be found on the Google AI website: https://ai.google/research/
Top comments (0)