Kristian T

👨‍💼 co-founder of 🥷 Mobile + iOS + ML + AI (on-device edge AI/ML) 🇦🇺 Sydney, regular presenter at CocoaHeads

[Kristian T @KristianTT]

promptfoo red teaming is legit, managed to pick up some nice finds in the adversarial tests

[Kristian T @KristianTT]

@arlogilbert @0to1ping @jonoringer Absolutely no way that’s parity with frontier models over cloud inference.

Also don’t be sucked into believing every benchmark, many are benchmaxing / overfitting - and real world performance is way lower compared to frontier.

[Kristian T @KristianTT]

@ariaradnia Be cautious of value traps

[Kristian T @KristianTT]

@waterboystocks It’s not just AI that’s eating them. I used to use Lightroom all the time for my Sony camera rig, bought a Fuji X100VI 2 years ago and haven’t touched Lightroom ever since, and couldn’t be happier about it.

[Kristian T @KristianTT]

@turbopuffer Jokes on you, I was going to use it anyway!

[Kristian T @KristianTT]

With Cursor coming onboard to SpaceX I wonder what the next bolt-on would be to make it a more holistic ecosystem. My bet would be on something like Vercel, for a tighter loop of build + deploy

[Kristian T @KristianTT]

@wohwki

[Kristian T @KristianTT]

@0xSero Quants available for us 5090 plebs?

[Kristian T @KristianTT]

Claude still has the best /loop mechanism by far, its crazy what you can achieve with a well refined objective and a locally scheduled loop

[Kristian T @KristianTT]

Was using WISRP Flow a tonne over the past handful of months, had awesome productivity from it, but found myself blabbering on too much to the agents. Typing is more refined, my brain seems to do an extra filter process before typing, whereas brain to speech has little filter

[Kristian T @KristianTT]

@StopBrokeNow @unusual_whales It's confirmed for the chip workers, over 25B total.

[Kristian T @KristianTT]

@calvinfroedge You misread how the markets work. Markets at the macro level go up and down on changes to liquidity. The thing with geopolitical issues / bswans is they can be worked around with liquidity injection. Don’t think about the headline - think “will liquidity increase or decrease?”

[Kristian T @KristianTT]

@thsottiaux would love a better UX in Codex(mac) when working with automations with high frequency runs. Digging into each of the "Previous runs" one by one is a bit painful. Would also love memory/smarts around how I create automations - that modal seems to always reset settings

[Kristian T @KristianTT]

@brettmacro The consumer is what was holding up the economy in the past, but not anymore - now economy is perked up by excessive government spending, and corporate capex. Consumer can be hurting, but currently that doesn’t matter. For now…

[Kristian T @KristianTT]

@gabrielchua Would have loved to be there!

[Kristian T @KristianTT]

@jonathan_wilke @cursor_ai Agreed. I currently have subscriptions / access to just about everything, including a beastly local-AI setup. Cursor is what my hands automatically reach for - but the whole ecosystem actually with their cloud agents and bugbot is really solid

[Kristian T @KristianTT]

@gabrielchua Would have loved to come along, this week didn't work out for me though. Hopefully next time!

[Kristian T @KristianTT]

@chribjel People need to remember the harness is also super important, esp for coding. Claude Code / Codex / Cursor are doing A LOT for the agent orchestration and overall result quality and experience. When jumping to the local AI ecosystem need an equally solid harness setup

[Kristian T @KristianTT]

@neural_avb Just hire a cloud GPU for a handful of hours (from an aggregator like vast AI) - will cost you $10-$20 to get a general feel on how good it can get, before forking out to invest hard in your local setup

[Kristian T @KristianTT]

@nlevnaut @LottoLabs In real world engineering cases you’re dealing with a skilled practitioner giving models clear bite sized tasks with guardrails. we are very much at the stage of “good enough” for those type of workflows.