There’s a new LLM contender in town

Sign up for the Daily Newsletter, sent exclusively to our premium subscribers. We break down the big and messy topics of Asia’s tech and startup community. Get the newsletter in your inbox everyday with a premium subscription.

Hello reader,

One thing that I’ve always found slightly irritating was how navigation apps would usually mispronounce locations in Singapore. Take the one used in my mum’s car for instance. It would completely mangle certain words to the point they were unrecognizable.

Kallang (“Kah-Lahng”) would become (“Keh-lehng”) or something worse. Google Maps would be slightly better but still make numerous mistakes. It underscored the fact that even though these apps have really advanced tech and were built by some of the smartest people around, they would not be able to understand other countries and cultures on a deeper level.

I believe that this concept also applies to AI and large language models (LLMs), especially given that they’ve been rising in prominence recently. Today’s story takes a look at how AI Singapore is trying to tackle this issue with its own LLM, dubbed Sea-Lion.

Today we look at:

• An LLM built for Southeast Asia
• Tokopedia’s new CEO
• Other newsy highlights such as Shopee’s lawsuit getting dismissed and Paytm’s shares crashing down.

Premium summary

Hear me roar

Image credit: Timmy Loen

I’m really proud of Singlish, the so-called mangled version of English mixed with other local languages and dialects. I think it’s important to preserve these types of cultural quirks, especially in the face of rapid tech advancements, so the arrival of Sea-Lion is definitely welcome.

• Balancing the scales: One key reason why AI Singapore built Sea-Lion is because there’s a need for LLMs that are not solely aligned with Western culture. As such, the Singapore government recently committed US$52 million to further develop it.
• Dark horse: In a demo event, AI Singapore showed how Sea-Lion fared compared to big-name LLMs like Meta’s Llama 2, Alibaba’s SeaLLM, and OpenAI’s GPT-4. The demo showed that Sea-Lion outperformed those models in terms of speed, accuracy, and succinctness.
• Hand in hand not head to head: That said, Sea-Lion is not meant to compete against these LLMs, according to AI Singapore. Rather, it’s designed to complement them by specializing in questions made in Southeast Asian languages.

Read more: Why Singapore’s LLM isn’t sweating GPT-4

News spotlight

Double Ts

Join Tech in Asia’s inaugural conference in Ho Chi Minh City on May 30

Quick bytes