Google Bard‘s updated language model makes it a math monster
Engineers have taught the AI math and reasoning
Since the advent of mainstream conversational AI with ChatGPT, Google has been eager for a slice of the action. Well, Bard is here now, and continues growing exponentially. Like its OpenAI-led rival, Bard has produced laughable responses to basic arithmetic and logical reasoning questions. However, Google has been developing another language model separately, which has now been fused with Bard to help with logic capabilities and even pave the way for autonomous code generation in the future.
If you aren’t familiar, Bard is based on Google’s Language Model for Dialogue Application (LaMDA), and as the name suggests, it is better at holding conversation than being your pocket logician. In 2021, Google announced its long-term vision for a new Pathways model, and in 2022, introduced the Pathways Language Model (PaLM). At the time, PaLM was a 540-billion-parameter language model trained using Pathways. Now, Google is fusing PaLM with Bard (via 9to5Google).
PaLM’s logical capabilities include arithmetic, code completion, semantic parsing, summarization, logical inference, common-sense reasoning, pattern recognition, translation, an understanding of physics, and even joke explanation. Here are a few examples Google shared last year, demonstrating these capabilities Bard lacked until recently:
In his tweet announcing the PaLM addition, Google staffer Jack Krawczyk says Bard is now better equipped to respond to multi-step word and math problems, and the code generation capability is coming soon. To be clear, ChatGPT has code completion capabilities, but isn’t polished enough in other aspects to provide reliable answers to logical and numerical questions. The bot also appears rather gullible and prone to coercion.
Hope you’ve had a chance to start testing out Bard! We’re constantly working on developing new features for it - today I wanted to share that we’ve improved Bard’s capabilities in math and logic by incorporating some of the advances we’ve developed in PaLM. /1 pic.twitter.com/RNMC5S9cRU— Jack Krawczyk (@JackK) March 31, 2023
Talking about ChatGPT with the New York Times, Google boss Sundar Pichai admitted Bard is in very early stages of development. He clarified Bard’s current test is to see the kinds of questions thrown at AI, which would help Google identify advanced models (like PaLM) which can be plugged in down the road.
With enhanced logical prowess, Bard could become every student’s assignment completion accomplice and a ready solver for complicated questions one encounters in real life. However, we suggest you pit Bard against ChatGPT regularly to find the more reliable AI, as their models continue evolving rapidly.