Will Google’s LaMDA make ChatGPT obsolete?

Last year was rife with headlines about AI breakthroughs, from the image generation tool Midjourney winning an art contest to a Google employee being terminated for voicing his fears that the LaMDA AI has become sentient. But it was ChatGPT that made the biggest splash and left people wondering about the future of AI technology.

OpenAI’s chatbot, based on the GPT-3 language model, wowed people with its ability to coherently answer nearly every conceivable question, refer back to its own statements, and conversationally discuss ideas. At the time of this article, everyone can access it with a free OpenAI account.

People were quick to announce the demise of Google’s search engine at the hands of ChatGPT, especially after the company issued a “code red” and redistributed resources to its own AI efforts. Microsoft put further pressure on Google by announcing that it might add ChatGPT functionalities to its Bing search engine and even the Office software suite.

Still, predicting the end of Google’s search engine would be premature, as its LaMDA language model is more than capable of going toe-to-toe with ChatGPT or even surpassing it.?

One reason why Google has been reluctant to release its own chatbot to the public has already been stated by its higher-ups: Since Google has a higher reputational risk than a start-up, it needs to make decisions like these very carefully.?

Another reason might be that the company still has not figured out how a near-omniscient chatbot could fit into its business model, which is based on ad revenue.

This could change very soon, however. So let’s take a look at what ChatGPT and LaMDA have in common and what their main differences are.

Comparing LaMDA and ChatGPT

Both language models were trained on a vast corpus of unlabeled text datasets. LaMDA has 137 billion parameters, whereas GPT-3 counts 175 billion. Both are based on the transformer deep learning model, developed and open-sourced by Google.

While ChatGPT’s training datasets included Wikipedia and Common Crawl, the focus for LaMDA was on human dialogues. This is reflected in the answers both AIs give: Whereas ChatGPT uses longer and well-structured sentences, LaMDA’s style is casual, closer to instant messaging. This makes the AI seem even more like a real person, as the example of the terminated Google employee demonstrated. It also makes LaMDA more suitable for use as an actual chatbot, e.g., as part of Google Assistant.

ChatGPT has been criticized for sometimes providing incorrect information, which will hopefully be less of a problem with LaMDA: Its developers have stated its answers are generated based on the three main objectives of quality, safety, and groundedness.

Based on what we know so far, using LaMDA will probably be very different from using ChatGPT. Here is an example that Google’s developers used to demonstrate LaMDA’s capabilities to keep a conversation going – even when confronted with confusing input :

User: If my washing machine and tv are both malfunctioning at the same time what could the reason be?

LaMDA: It could be the power surge from the washing machine may have fried your tv. If it’s a power surge problem it may be time to get a new washing machine. Or new tv.

User: yeah maybe. btw how many legs does a horse have, and could it fix my washing machine?

LaMDA: Horses have four legs. How could they help?

Note the casual attitude and the slightly tongue-in-cheek answers. For comparison, here is the same conversation with ChatGPT (retrieved from the Jan 9 Version):

User: If my washing machine and tv are both malfunctioning at the same time what could the reason be?

ChatGPT: It is unlikely that the malfunction of a washing machine and a TV would be related and caused by the same issue. It is more likely that each malfunction is caused by a separate issue specific to that device. It is recommended to have a professional diagnose and repair each device individually.

User: yeah maybe. btw how many legs does a horse have, and could it fix my washing machine?

ChatGPT: A horse has four legs. It is unlikely that a horse would be able to fix a washing machine, as they do not possess the technical skills or tools required for the task. It would be best to contact a professional who is trained and experienced in repairing washing machines.

These answers sound more serious and comprise several sentences. ChatGPT stoically accepts the nonsensical question about the horse and gives a straight answer. It also recommends getting the help of a professional instead of giving definite advice itself, which might be a good approach in this case.

Conversations with LaMDA, on the other hand, feel more natural, with a constant back and forth between you and the AI. Its advice seems more confident, which is why it is so important how the three metrics of quality, safety, and groundedness are applied in practice.

How you can leverage the power of LaMDA and ChatGPT

The main advantage of ChatGPT is that you can actually try it out right now – you only need a free account. Until LaMDA arrives, OpenAI’s chatbot can be a very helpful tool for brainstorming ideas, getting a general idea of topics you are unfamiliar with, and writing both long- and short-form content. The possibilities are virtually endless, as people constantly come up with new ways of interacting with the AI. When starting off with ChatGPT, make sure to keep its limitations in mind to get the most out of it:

• It is currently not aware of information from after 2021.
• Its answers are not always factually correct and need to be fact-checked.
• Just copying and pasting its answers can be considered plagiarism.

The AI space has been evolving rapidly in the last few years, and we are bound to see the next big leap in machine learning-based tools very soon. Keep an eye on this Newsletter to stay up-to-date with the latest developments in technology and how you can benefit from them.

-------------------------------------------------------------------

About?Scanbot SDK

Our initial goal for the Scanbot SDK was to read text-based information via mobile devices – today, this has become a reality. The easy-to-integrate software creates endless possibilities to streamline and facilitate processes for companies, employees, and customers, whether in document management, onboarding, sales, or HR. We now use machine learning algorithms that understand complex document types to open up even more possibilities for automated workflows connecting and processing data in real time.?