One of many largest bulletins of Google I/O was the disclosing of PaLM 2, Google’s newest and most superior giant language mannequin (LLM). With the transition to PaLM 2, Google claimed Bard would have enhanced logic, math, and coding capabilities.
Right this moment, it was time to place these talents to the check.
Additionally: I examined Google Bard’s latest coding expertise. It did not go effectively…once more
Since Google Bard now not has a waitlist, to get entry, all it’s a must to do is go to Google Bard’s web page and signal into your private Gmail account.
As soon as I did that, I used to be prepared to begin asking away.
The primary noticeable change was that the Bard had the choice of darkish mode, one of many options Google revealed at I/O meant to assist cut back eyestrain when utilizing the chatbot. This was a optimistic signal.
Additionally: The way to use Google Bard now
Subsequent up, it was time to substantiate the chatbot’s largest improve, PaLM 2. To confirm the upgrades for myself, I requested the chatbot “What LLM are you operating below?”. The response was surpising.
Regardless of Google’s announcement at Google I/O, Bard instructed me it was nonetheless operating on LaMDA.
I’m operating below the Google AI LaMDA language mannequin. LaMDA stands for “Language Mannequin for Dialogue Functions”.
Screenshot by Sabrina Ortiz/ZDNET
After I adopted up that response by asking, “I assumed you had been operating on PaLM 2”, I used to be met with the same response.
I apologize for the confusion. I’m nonetheless below growth, and I’m not but capable of run on PaLM 2.
Screenshot by Sabrina Ortiz/ZDNET
Is the chatbot hallucinating and offering incorrect responses because it has up to now? Or may the chatbot not been switched to PaLM 2 but, as Google declared yesterday? I reached out to Google for some readability, however the response left me with as many questions as I needed to start with.
“LLMs (Bard included) can hallucinate and current inaccurate data as factual,” stated the Google spokesperson.
Additionally: Each main AI characteristic introduced at Google I/O 2023
Bard’s lack of ability to precisely say what LLM it’s operating on may merely be a hallucination. Nonetheless, one of many largest targets of utilizing a extra superior LLM for Bard was to lower the variety of hallucinations that occurred.
Regardless of how superior Google claims PaLM 2 is, to date, Bard appears to proceed to be affected by the issues of the previous together with restricted coding talents. Maybe Google overestimated PaLM 2’s capabilities.