Brian Bensch

Founder and owner @snowschoolers, marketing & growth advisor. Prev @Google @Airbnb @Hipcamp

[Brian Bensch @bbensch]

@tahoe_tech_bro yah why you charging 4X?!?!

[Brian Bensch @bbensch]

@tahoe_tech_bro your pricing is dumb

[Microsoft Azure @Azure]

Kimi K3 is now available on Microsoft Foundry through Fireworks AI. It's a 2.8 trillion-parameter open-weight model from Moonshot AI with a 1 million-token context window.

Learn more: https://t.co/7JQxIES0MI

[Brian Bensch @bbensch]

2006: anyone can have their own online blog
2016: anyone can learn to code in ~4 months
2026: no one learns to code they just prompt AI and it's basically free.

[Kunal Kushwaha @kunalstwt]

You have to check this out. It’s impressive that you can use the most powerful open source model for coding for basically “free”.

I tested the new Kimi k2.5 model on @FireworksAI_HQ and it is 90% cheaper than the big names. It costs $0.60 per million input tokens compared to $5.00 on Claude Opus. Fireworks is the official launch partner for Kimi K2.5, and runs more than twice as fast as other API providers based on Artificial Analysis at ~200 tok/s.

I put it to a real test by asking it to build a full React Super Bowl betting app. It handled everything from logins to live leaderboards by self-directing a swarm of 100 sub-agents. It finished the task 4.5x faster than a normal setup.

You can also turn screenshots/videos into websites. I gave it a screenshot of a pro sports site and it wrote the React code to match the look perfectly.

Try Kimi on Fireworks, the sponsor of this post, as they’ve made it production-ready. It also supports full parameter RL tuning. This means you can actually customize the model to your specific product and end up with quality that exceeds the closed-source giants.

You’re getting the speed of a specialized platform with the brain of a frontier model, all while saving ninety percent on your inference costs.

👉 It feels free to run and you can experiment more: https://t.co/wz0f6GZqqa

[Brian Bensch @bbensch]

hey @tahoe_tech_bro here's an unpopular opinion: Moltbook isn’t the "emergence of synthetic culture," it’s just the world’s most expensive improv class for GPUs. 🦞
It’s not AGI. It’s a chatroom for your API bill.

[Brian Bensch @bbensch]

I wish @tahoe_tech_bro would check his twitter more ;-)

[Brian Bensch @bbensch]

@JohnnotJon Airpost 🙏

[Lin Qiao @lqiao]

🚀 Eval Protocol is Open Sourced!

Reinforcement fine-tuning is complicated, because there are hundreds of environments and tens of trainers you can pick and choose and integrate with. Even worse, in production, agents don’t live in clean “gyms.” They operate in messy, async environments - flaky APIs, partial observability, conflicting objectives, long feedback loops.

We solve that problem by open sourcing Eval Protocol. The goal is for you to build your production RFT flow without reinventing the wheel of managing such complex integration.
👉 Day 0 support for trainers and environments like TRL (@huggingface), @rllm_project , OpenEnv (@PyTorch),  as well as support for proprietary trainers like @OpenAI RFT and @thinkymachines Tinker. More to come.
👉 Instrument agents in production instead of toy or simulated environments
👉 Move from offline benchmarks to live, continuous improvement

[Brian Bensch @bbensch]

@powderski I ski every day of the season until I sometimes get bored. Too much better on two planks than one, that I can explore whole mountain and push myself. When i'm bored and conditions are meh, or have friends in town, then I practice my riding.

[Brian Bensch @bbensch]

@StormSkiJournal @MtRoseSkiTahoe @Sugarbush_VT @SolitudeMTN hot take: it's great for locals and an obvious good idea for any independent resort that still wants to sell more day tickets. Much less so for any megapass mountain that derives 70%+ lift revenue from pass sales and is more focused on destination travelers than locals.

[Brian Bensch @bbensch]

@i80chains did the feed break I presume?

[Brian Bensch @bbensch]

@liftblog Best case scenario that the majority of folks can root for is this: new ownership will invest $5 to $10 million toward their objective building a mini version of Yellowstone club and not give up until the new lifts are in . Once the project fails someone else like MCP swoops in

[Brian Bensch @bbensch]

@OpenSnow team -- looks like something is broken with the data sources in the AQI forecast map layer. figured I'd flag if anyone is able to take a look this week. thx! https://t.co/UPlZc2mkoQ

[Brian Bensch @bbensch]

Am I crazy to think that homeowners insurance might just cease to exist in California a decade from now? https://t.co/PB2nzrfvTY

[Brian Bensch @bbensch]

@matt_lorelli Do you know if that’s Lake Mary in the midground?

[Brian Bensch @bbensch]

@CALFIREAEU You probably want to go ahead and add that link to your Twitter bio 😊
E.g.

[Brian Bensch @bbensch]

@i80chains anyone excited as I am about this storm to want to bet how many hours I-80 sits closed between now and Sunday evening? there are 88 hours between 8am Thurs and midnight Sunday. I'll set my line at highway being closed for 72 of said hours...over or under?

[Brian Bensch @bbensch]

@HemlockWhiskey @liftblog @forestservice if anyone here has had success filing a FOIA (freedom of information act) request with the USFS would love to hop on a quick zoom call and get some advice. went down rabbit hole a few weeks back trying to do so but gave up.

[Brian Bensch @bbensch]

@tejones Crazy how there was zero coverage or commentary about that play.