동욱

AI forward deployed engineer. I run 40+ Hermes Agent cron jobs and post how they're set up and what breaks. Made kstack. Korean, on F1 in the US.

[동욱 @dwkim601]

https://t.co/Aj1OKgAyhW

[동욱 @dwkim601]

@mercor @jake_cole1 Kernels and aircraft design are in one score. I can't tell which kind of work failed from that.

[동욱 @dwkim601]

@badlogicgames Swipe to archive is the demo. The row comes back if Gmail refuses.

[동욱 @dwkim601]

@JaynitMakwana Closing the laptop mid-task and a new computer are different problems.

[동욱 @dwkim601]

@philhchen Products were on default settings. The models got a harness with native search and reading tools. Not the same setup.

[동욱 @dwkim601]

TIP: A CLI chat session does not automatically fire cron jobs. The gateway ticker ticks every 60 seconds. hermes gateway start runs that service. hermes cron tick never starts a gateway, and with none running it skips agent jobs.

[동욱 @dwkim601]

NEWS: I'd watch the cheap tier, not the headline. A bigger model beats turning effort all the way up, until the bottom, where every step up costs more.

[Artificial Analysis @ArtificialAnlys]

With the current Claude lineup, moving up a model tier beats maxing out effort to achieve the highest intelligence per dollar

At nearly every Cost per Task, we see the largest model sits on the performance vs. cost Pareto frontier for @AnthropicAI models:

➤ Claude Opus 5.5 (xhigh, 57) outscores Claude Sonnet 5.5 (max, 55) with 40% lower Cost per Task and about half the Time per Task
➤ Claude Fable 5.2 (high) beats Opus 5.5 (max), with smaller cost and time savings, and Fable 5.2 (low) beats Opus 5.5 (high)
➤ This pattern doesn’t hold at the cheapest tier, where every step up from Claude Haiku 5.5 (max) costs more

[동욱 @dwkim601]

@HermesAgentTips right as the hackathon was ending, the catalog went past 2,000. a jump like that is not the same as a review.

[동욱 @dwkim601]

@Voxyz_ai One component at a time has an end. The prompt never says how you'd know nothing broke.

[동욱 @dwkim601]

@d0znpp not being allowed to generate text is the bet. the cost line leans on 'output tokens free'.

[동욱 @dwkim601]

Source: https://t.co/nI6LFm0R8p
https://t.co/r6PFvBJofN
https://t.co/9ZTNgGjtsI

[Artificial Analysis @ArtificialAnlys]

Google has released its first decision model, Gemini Decision 1, scoring 66 on the Artificial Analysis Decision Index and leading in knowledge and reasoning benchmarks

Gemini Decision 1 from @GoogleDeepMind scores 66 on the Artificial Analysis Decision Index v2.1, in line with Elephant 9B (@baselabs) and Jev 1.14 (@typesafeai), and 7 points behind leaders Drex Max v1.1 (@NaceAI) and GLiDE v1.2 (@fastinoAI), which both score 73. It is priced at $0.10 per 1M input tokens and costs $345 per 1M Tasks across the Index, around three times higher than Drex Max v1.1 ($103) due to its higher pricing

Key results for Gemini Decision 1:

➤ Not yet frontier on Decision Index: Gemini Decision 1 scores 66 on the Artificial Analysis Decision Index. It leads knowledge and reasoning benchmarks, scoring 66 on AA-Omniscience-Decision and 49 on HLE-MultiChoice. It trails on other evaluations, including our new proprietary classification and extraction benchmark AA-Paperclip, where it answers all 15 questions correctly on 33% of documents vs. 63% for Drex Max v1.1

➤ More than 3x the Cost per Task of leading models: Gemini Decision 1 costs $345 per 1M Tasks, compared with $103 for Drex Max v1.1, $156 for Jev 1.14. This difference is driven by its token price of $0.10 per 1M input tokens vs. $0.03 for Drex Max v1.1. Gemini Decision 1 has a similar Cost per 1M Tasks to GPT-6 Luna (Decisions, $339)

➤ Strong calibration for reasoning tasks: We measure calibration with the Brier score, the average squared gap between the probability a model assigns to its chosen answer and whether that answer was correct (lower is better). Gemini Decision 1 has a Brier score of 0.18 across the Index, between Drex Max v1.1 (0.16) and Jev 1.14 (0.19). Calibration varies by task. with its Brier score of 0.20 on HLE-MultiChoice among the lowest of all models in the Index, while its 0.14 on When2Call compares with 0.04 for Drex Max v1.1

[동욱 @dwkim601]

NEWS: The launch post calls Decision 1 frontier. Artificial Analysis says it isn't frontier yet on their index, and that it costs more per task than Drex Max v1.1.

[동욱 @dwkim601]

@Da7_Tech A broken page is a thin base for calling every GPT model worst at taste.

[동욱 @dwkim601]

@davidvgilmore Open source means the router can be read. The Opus line still doesn't say which tasks.

[동욱 @dwkim601]

@ArtificialAnlys @joinHandshakeAI Calling it tools-off leaves out the other change. That run is one-shot, so the thinking and code steps are gone too.

[동욱 @dwkim601]

@Lummox_eth a truck on its roof isn't a model ranking.

[동욱 @dwkim601]

@hanakoxbt reading the diff can still miss a test nobody ran.

[동욱 @dwkim601]

@andonlabs Second place on drawing floorplans isn't the same as matching a person.

[동욱 @dwkim601]

@yuhasbeentaken that 10x on Claude Pro still depends on the model and how you use it.

[동욱 @dwkim601]

@mitansh_j07 @NEARProtocol a freeze can't undo a payment that was already broadcast

[동욱 @dwkim601]

Budget Broker MCP, built at NEAR HACKATHON. Several agents spend from one NEAR testnet wallet and none of them holds the key. In the testnet demo, one agent paid, one went over its limit and was denied, and a frozen one was denied. Denied requests are never signed.

@NEARProtocol #NEARHackathon

[동욱 @dwkim601]

@IBuzovskyi @NousResearch that one setting isn't the install. enable the plugin, then restart the gateway.