David Corbett

Chief Product and Technology Officer, lover of great coffee.
Husband of one, Father of two boys, 2 cats, and 2 international students.

[Paul Couvert @itsPaulAi]

Friendly reminder that Google has an official app to run Gemma 4 on your phone.

- 100% open source
- Fully offline and private
- Multimodal with text/audio/image
- Works with Gemma E4B and E2B

And the app is available on both iOS and Android.

Steps and download below

[David Corbett @101shipit]

@karlprosser 2 Claude Max plans, Gemini Enterprise, cursor pro, amp payg

[David Corbett @101shipit]

If only I had a token for every time I heard, "I've run out of tokens"

[David Corbett @101shipit]

@_philschmid @antigravity I think it depends on where you are doing the work.
If it's on your laptop with all your other private stuff like bitcoin wallets etc then it's incredibly risky. But if you are working on a gcp/aws vm where the only things there are the dev tools to do the work then its more safe

[Google DeepMind @GoogleDeepMind]

We’re using AI to work on root node problems – fundamental scientific challenges that unlock societal benefits. 🧪

From fusion and superconductors to entirely new materials, our CEO @DemisHassabis discusses what comes next after #AlphaFold – all on our podcast with @fryrsquared. ↓

Timecodes:
01:42 2025 progress
05:14 Jagged intelligence
07:32 Mathematical version of AlphaGo?
09:30 Science vs commercialization
12:42 Scaling
17:43 Genie and simulation
25:47 Evolution in simulation
28:26 AI bubble
31:56 Building ethical AI
34:31 AGI
44:44 Turing machines
49:06 How it feels to lead

[David Corbett @101shipit]

@NZAA what is up with with this. Your login to my aa insurance site is broken

[David Corbett @101shipit]

@autohiveai Excited for the Autohive launch.
It's a super cool way of creating agents to help you automate your business.

[John-Daniel Trask @traskjd]

Do you share aircraft time with others?

Astroskiff is designed for aviation groups to easily manage their syndicate.

Designed to streamline aircraft bookings, invoicing and maintenance management. Built by passionate Kiwi pilots, for aviators everywhere. 👇

[John-Daniel Trask @traskjd]

Check it out here: https://t.co/6fEZ6getRG

It's just launched so the creator is looking for feedback, and hopefully some customers. I don't fly but wanted to share this with aviation community.

[David Corbett @101shipit]

@AstroskiffATC So excited for this
https://t.co/XzqBiTTYPM

[Astroskiff @AstroskiffApp]

We're excited to announce https://t.co/pMGreAd9dh is taking flight! Designed for aviation groups wanting to streamline scheduling, communication, logbooks, invoicing, and group operations.  Take your group flight operations to new heights 🛫🛩️

[Jim Fan @DrJimFan]

We are living in a timeline where a non-US company is keeping the original mission of OpenAI alive - truly open, frontier research that empowers all. It makes no sense. The most entertaining outcome is the most likely.

DeepSeek-R1 not only open-sources a barrage of models but also spills all the training secrets. They are perhaps the first OSS project that shows major, sustained growth of an RL flywheel.

Impact can be done by "ASI achieved internally" or mythical names like "Project Strawberry".
Impact can also be done by simply dumping the raw algorithms and matplotlib learning curves.

I'm reading the paper:

> Purely driven by RL, no SFT at all ("cold start"). Reminiscent of AlphaZero - master Go, Shogi, and Chess from scratch, without imitating human grandmaster moves first. This is the most significant takeaway from the paper.
> Use groundtruth rewards computed by hardcoded rules. Avoid any learned reward models that RL can easily hack against.
> Thinking time of the model steadily increases as training proceeds - this is not pre-programmed, but an emergent property!
> Emergence of self-reflection and exploration behaviors.
> GRPO instead of PPO: it removes the critic net from PPO and uses the average reward of multiple samples instead. Simple method to reduce memory use. Note that GRPO was also invented by DeepSeek in Feb 2024 ... what a cracked team.

[David Corbett @101shipit]

@kellabyte +2
Inline process calls rather than service calls over the network for basket promotion calculator for the first one purely for speed.

[David Corbett @101shipit]

Would we get better results from AI if they had the ability to act like a toddler and just constantly ask why?

[Corey Quinn @QuinnyPig]

Their financial guidance for next quarter makes no mention of the fact that they just flipped a switch that will boost AWS earnings by $3b-5b a year in the form of charging per IPv4 addresses. Can't wait for someone to be surprised by that and claim it's new sales.

[Melanie Bracewell @meladoodle]

Here is a joke that only makes sense to New Zealanders so I’ve posted it to my international audience.

[Allie K. Miller @alliekmiller]

The best 20 seconds of AI video I’ve seen today 😂

[Laura Bell Main @lady_nerd]

What would #appsec look like if it was inclusive? Let me show you.

Today @safestack makes another step forward in our mission of bringing security to #software teams, big and small, worldwide from just $10 per month.

Want to make #appsec part of your dev world - let's go!

[Tom Knowles @tkbeynon]

There’s this guy on TikTok who does nursery rhymes in the style of his favourite bands and the production standards are insane. This Mumford and Sons style rendition of You Are My Sunshine is genuinely as good as a song they would release

[Pete @nonmayorpete]

Stanford just released a MASSIVE 386-page report on the state of AI.

Here are the 12 most interesting trends that you should know.

[0xDesigner @0xDesigner]

Probably the best thing you'll see today.

In 2017, a group of developers hilariously competed for who could create worst volume control interface in the world.

The results 🧵

1/22