Oli Gaymond

@Android Intelligence Product Lead

[Oli Gaymond @ogaymond]

@suchenzang With LLMs time to first prototype is low but time to full product remains approximately the same. So you should iterate through more prototypes and increase quality/robustness/value of the product you end up with.

Just don't mistake the prototype for the product :)

[Oli Gaymond @ogaymond]

@suchenzang Been thinking about this more generally. Major repeatability issue in adopting AI is the people doing the adopting. People look at how frontier labs use AI day to day and think they will be same in a few months...perhaps only if they hire the same people as the frontier labs

[Oli Gaymond @ogaymond]

The question remains: is AI value accretive or value destructive?

The value of the answer is now in the trillions.

[Oli Gaymond @ogaymond]

Excited to make Gemma 4 available on Android in the AICore Developer Preview in partnership with @Arm

[Arm @Arm]

Real-time assistance. Seamless communication. Greater personalization. On-device AI with Gemma 4, built on Arm. 💪

[Oli Gaymond @ogaymond]

Thank you @Qualcomm for your great partnership. Excited for Android developers to experience Gemma 4 on @Snapdragon

[Qualcomm @Qualcomm]

With day‑zero enablement on @Snapdragon, we’re delivering Gemma 4 to developers early, unlocking fast access to next‑gen on‑device AI. Congrats Qualcomm and @googlegemma teams on moving open‑source AI forward.

[Demis Hassabis @demishassabis]

Excited to launch Gemma 4: the best open models in the world for their respective sizes. Available in 4 sizes that can be fine-tuned for your specific task: 31B dense for great raw performance, 26B MoE for low latency, and effective 2B & 4B for edge device use - happy building!

[Google DeepMind @GoogleDeepMind]

Meet Gemma 4: our new family of open models you can run on your own hardware.

Built for advanced reasoning and agentic workflows, we’re releasing them under an Apache 2.0 license. Here’s what’s new 🧵

[Oli Gaymond @ogaymond]

We have a lot coming this year to make building with AICore easier and more capable. These new models are a first part of that. We can't wait to get your feedback and build with you.

[Oli Gaymond @ogaymond]

Congrats to @o_lacombe and team for the Gemma 4 launch today. A range of open models with leading capabilities.

https://t.co/BSlZj1m9Fu

[Oli Gaymond @ogaymond]

On all devices you can experience the quality of these new models with a CPU implementation. On flagship devices with Google, Qualcomm and MediaTek chipsets you can also experience the performance with early hardware accelerated versions.

[Oli Gaymond @ogaymond]

Been making these for the team. Think I should get you one @o_lacombe ?

[Oli Gaymond @ogaymond]

@mortenjust Something coming soon, but this is a slow burn thing.

[Oli Gaymond @ogaymond]

So much work to be done on interaction primitives.

[Oli Gaymond @ogaymond]

@AndrewOrlowski Brooks was right but talking about within a project. If more tasks are being done then LoC going up is fine.

AI coding mostly lowers barrier to entry meaning tasks that were previously uneconomic can now be done. It's like arrival of VisiCalc. It expands market.

[Oli Gaymond @ogaymond]

Lots more to talk about later this year as we build towards personal AI helping you throughout your phone.

[Oli Gaymond @ogaymond]

It's so annoying when you're talking to someone and you get cut off as they go through bad connectivity. The same is true of AI assistants. To be helpful they need to be reliable.

One of the foundations is high quality speech transcription. Enter ML Kit on-device speech API 💬

[Oli Gaymond @ogaymond]

Gemini Nano is a foundational model and enables a wide range of capabilities, all running directly on your device and available at all times. Great when you're mobile.