Google Brings a Powerful Multimodal LLM to Your Laptop - DTNS 5283
Daily Tech News Show
Tom Merritt
4.8 • 1.5K Ratings
🗓️ 4 June 2026
⏱️ 25 minutes
🧾️ Download transcript
Summary
We also talk about Nintendo’s replaceable battery in Europe, some hope for energy storage, and an explanation of Apple's smart glasses strategy as well.
Starring Tom Merritt and Jenn Cutter
Show notes found here.
Hosted on Acast. See acast.com/privacy for more information.
Transcript
Click on a timestamp to play from that location
| 0:00.0 | This is the Daily Tech News for Thursday, June 4th, 2026. |
| 0:14.0 | We tell you what you need to know, give you some context, and try to help each other understand. |
| 0:18.3 | Today, Google just released a powerful model you can run on laptops |
| 0:22.5 | with 16 gigs of memory. Yeah, so not, you know, not your Chromebook, but still, a lot of laptops |
| 0:28.8 | be able to run that model locally. It's an interesting release. We're going to talk about it. I'm Tom Merritt. |
| 0:33.8 | I'm Jen Cutter. Let's start with what you need to know with that big story. |
| 0:46.8 | So yeah, Gem Gemma, Gemma, Thore is the current open source models from Google. This is different than Gemini. It's a clever Gemma leads to Gemini. It's kind of like Android versus, you know, |
| 0:53.1 | the Google version of Android. But Gemma |
| 0:55.6 | 4 has released four models. Two of them, you pretty much need to run in a data center or a really |
| 1:02.7 | beefy desktop, like one of the ones you could buy from Nvidia or Microsoft. And then two of them |
| 1:08.0 | are meant for mobile, but they're not as capable because they're meant to run on very small machines. |
| 1:12.6 | Gemma 412B just got released now, and it can run on any laptop with at least 16 gigabytes of RAM or V RAM. |
| 1:21.6 | So like we said, not your smallest laptops. |
| 1:24.6 | And it is released under the Apache 2.0 open source license so you can get it for free |
| 1:30.0 | and do what you want with it but it is more capable than the two Gemma4 models previously |
| 1:35.6 | released for mobile and it could still be run locally without you having to buy a super expensive |
| 1:40.9 | hardware it has 11.95 billion parameters and uses half the memory footprint of the |
| 1:48.1 | higher model, 26BMOE, but benchmarks pretty close, almost as capable. And it has a rather large |
| 1:55.3 | 256,000 token context window. So you can put some beefy stuff into it, some big company manuals or such. |
| 2:03.3 | It can handle agenic workflows. It can do multi-step reasoning and it's multimodal using a new |
| 2:11.3 | approach called multi-token prediction drafters. That takes unused processing cycles to predict future tokens well enough that it can |
| 2:21.2 | speed things up and make it more power efficient. The multimodal stuff is also really impressive. |
... |
Transcript will be available on the free plan in 22 days. Upgrade to see the full transcript now.
Disclaimer: The podcast and artwork embedded on this page are from Tom Merritt, and are the property of its owner and not affiliated with or endorsed by Tapesearch.
Generated transcripts are the property of Tom Merritt and are distributed freely under the Fair Use doctrine. Transcripts generated by Tapesearch are not guaranteed to be accurate.
Copyright © Tapesearch 2026.