@scosman

Building Kiln AI. Formerly Apple Photos, ShoeboxApp, Msft. He/him

 Toronto
kiln.tech
Joined October 2008
  • Tweets3.1K
  • Following531
  • Followers530
307Media
Building AI agents shouldn't be whack-a-mole. Solve a problem once. Keep it solved. Kiln v1, launching today ↓
3
8
5
42
8,761
Steve Cosman retweeted
Olivia Chow
@oliviachow19h
Tonight, the people of Toronto have spoken. And you have given me the extraordinary honour of continuing to serve as your mayor. I am grateful. I am humbled. And I am so proud of our city.
590
650
103
6,788
91,831
Lies, damn lies and benchmarks. All in last 24 hours. Oh: we're building a benchmark btw.
1
54
Tonight! toronto.aitinkerers.org/p/ai…
57
Steve Cosman retweeted
Kiln AI@getkilnOct 13
Software fantasy > Software factories
1
3
57
Steve Cosman retweeted
Kiln AI@getkilnOct 12
Eval Builder V2: building AI evals has never been this easy (and accurate!)
1
2
68
We need a NO_AGENT tag for emails. I want to be able to connect my inbox to agents, but not give it access to authentication codes.
73
Talk about what you make, loudly and often. I shipped a "build videos in Claude" framework back in May. No one looked. Now it's all everyone is talking about.
Steve Cosman@scosmanMay 21
How we made our launch video with Claude Code (and open-sourced the tool) kiln.tech/blog/we_made_our_l…
1
1
159
Excited for this one: an eval builder that works with you. It does the heavy lifting. You bring the judgment. Free at kiln.tech
1
1
1
163
He's not kidding
nathan fielder
@nathanfielderOct 7
Oct 16
36
It's only sovereign AI if it comes from the Sovereign region of France. Otherwise it's just sparkling intelligence.
Mistral AI
@MistralAIOct 6
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding. • Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure. • Available to all via API today. Working with cybersecurity partners privately. Open weights release end of October.
2
55
The trial of Opus 5
36
I found this funnier than my team did
🤖 Made with AI
1
5
102
Opus 5: you know what you did.
7
AI a year ago: you hit tab 140k times! AI Now: to verify this change I installed GNOME, Plasma, AppIndicator and more in a VM, and visually checked the rendering of each one. I also patched a bug I found Debian.
33
Steve Cosman retweeted
Nathan Lambert
@natolambertSep 22
Sir what is going on here
Claude
@claudeaiSep 22
Replying to @claudeai
At its default effort setting, Opus 5.5 delivers frontier results for a fraction of the cost per task, often beating other models running at their highest settings. It also generates output more than 30% faster than Opus 5.
176
96
55
3,636
636,155
Using autoresearch on Jev. Once again, harness optimizations beat model optimization. kiln.tech/blog/auto_optimizi…
2
116
Github Issues Honeypot: an open issue on popular repo that seems simple but is nearly unsolvable. See how many "people" send AI patches in a week.
2
73
it's been done
53
I made a better Granola, because why not. Local, private, offline, Mac native and open source. biscottiapp.com
1
1
12
202
Everyone, play along... please.
🤖 Made with AI
78
Load more