Kevin Simback 🍷

The great de-coupling - one of the best pair trades of the AI cycle Long memory manufacturers / short non-AI memory buyers

I recently setup a Hermes agent for my family, put us in a WhatsApp group The best feature so far - logging into the school portal every day and sending announcements, grades, homework, etc. Now working on summer travel planning Total game changer, dunno why I didn’t do this sooner

Sometimes the math maths even when it doesn’t

@Hiraweb3 🙏

@KSimback love seeing the breakdown of a crowded space getting clarity

Inference demand is going to continue to explode, so I wanted to take a look at every inference provider I could find 70+ providers researched -> 26 featured in this article I thought it would just be a commodity service, but there's a lot that differentiates them

@mikhryc0x I enjoy the process

@KSimback researched 70+ providers so we don't have to, respect the sacrifice this man ate the whole menu so we could just order

@m4xim1l1an @SambaNovaAI Thanks will check this one out

@KSimback @SambaNovaAI in the speed king section is missing

@AntSeedAI Thanks, will take a look. Have not added any p2p inference services but following some of the projects like Darkbloom, Hyperspace and a few others

@KSimback Give AntSeed a try and see if you should add it….

@vmiss33 Thanks for the shout!

Great write up of providers out there

@katie_haun Congrats on the raise!

@EtherCoins @Teknium Ok let me know when there’s a more accurate one

@KSimback @Teknium Is inconsistent enough to advice you not to. I will prepare a v0.1 that will be much more accurate.

I have to go out of town for a funeral thru the weekend but I am leaving everyone with one new cool feature inspired by ralph loops and Codex's upcoming /goal feature. If you use /goal , it will start a loop with a supervisor model determining whether the task completed at the end of an agent loop - if it hasn't it will force it to keep going until it's done! Enjoy and have a great weekend. PR: github.com/NousResearch/h…

@matlabs0 Not that I’m aware of

@KSimback That’s actually a game changer… If we can abbreviate the output/format in the .md container’s it will save time Is there a cheat sheet to look at?

I thought I knew Claude well, but I just today learned there are some shorthand “secret” codes 1. Put “L99” at the end of a prompt and it makes Claude respond at “Level 99” maximum expert depth and reduces hedging and forces committed recommendations 2. Put “OODA” at the beginning of a prompt and Claude will structure its response using the military OODA loop framework (Observe -> Orient -> Decide -> Act) 3. Put “SCAFFOLD” at the beginning of a prompt for Claude to generate a full step-by-step project setup or action plan There are a few others but those 3 I found most useful These are not official commands but apparently they work because Claude’s training data contains thousands of examples of people using them So the model has learned the patterns and adjusts its response style accordingly Pretty cool to know, if you’ve found other good ones drop them in the comments

@willchen500 @lex_node Meanwhile on LinkedIn…

Harvey and Legora are essentially sales organisations that resell tokens. They have hired legions of ex big law juniors and mid levels as sales people (“GTM”) along with some ex partners to wine and dine their former colleagues. They slap on a UI that makes them look different from ChatGPT but the product differentiation and vertical specific features are far and few in between. You could just as well use both for any white collar job. Their web apps are basically 1. A chatbot interface 2. A projects function where you can upload your files 3. A tabular review function where you can bulk review documents in a table 4. Workflows which are just custom prompts you write for the chatbot or tabular review. I was able to build everything plus some additional functionality they do not have like version control in mikeoss.com in two weeks. I call this the “token reseller theory”. They are like car dealers or real estate agents but for tokens. The model providers get them to do the selling to crack open the reticent legal market. What happens to H/L now that the model providers want the market for themselves? Does not bode well for them.

@dimzredemption OODA is good if you’re trying to make a decision with some level of uncertainty, it’s a battle tested framework L99 when you don’t want any BS

@KSimback the L99 one sounds like a panic button for when u need claude to stop being polite and just pick a side already OODA probably makes it think like a pilot or something

@so_ainsight @Teknium Appreciate the shout for Hermes Atlas, check it out: hermesatlas.com

セッションをまたいで記憶を保ち、経験から自分でスキルを作って育つAIエージェント Hermes Agent。ガチで使い倒すためのGitHubレポ10選： 21 Hermes Agent 本体 Nous Research公式のコアリポジトリ。MITライセンスで自由に使える。 github.com/NousResearch/h… 2. Hermes-Wiki Hermes Agentのソースコードを解説するコミュニティWiki。実装理解に効く。 github.com/cclank/Hermes-… 3. Atlas エコシステム地図。100+のツール・スキルを俯瞰でき、RAG検索にも対応。 github.com/ksimback/herme… 4. Control Interface セルフホスト型ダッシュボード。複数エージェント・長時間タスク・記憶を一画面で管理。 github.com/xaspx/hermes-c… 5. Skill Factory タスクを振り返って新スキルを自動生成・追加。エージェントが自分の武器を自製。 github.com/Romanescu11/he… 6. Maestro ローカル動作のマルチエージェント協調ツール。Codex・Claude Code・Geminiを横断して構造化記憶と引き継ぎを管理。 github.com/ReinaMacCredy/… 7. Hermes Agent Camel 信頼境界（CaMeL）を組み込んだフォーク版。本番運用の防護に向く。 github.com/nativ3ai/herme… 8. Hermes HUD TextualベースのTUI監視ターミナル。意識の流れとメモリ状態をリアルタイム可視化。 github.com/joeynyc/hermes… 9. Hermes Alpha クラウド環境向けのHermes Agentデプロイ用テンプレート。Makefileと設定例同梱。 github.com/kaminocorp/her… 10. Awesome Hermes Agent コミュニティ厳選のプラグイン・プロンプト・教材まとめリスト。 github.com/0xNyk/awesome-… 保存して順に試してみて。

@Futurenvesting Incredible

@KSimback It's truly mind-blowing growth

This is absolutely insane - Anthropic’s revenue run rate went from $30b at the end of March to $44b at the end of April They just added 3x Palantir’s entire annual revenue in ONE month Completely unprecedented

@ai_hyperbull Yeah that is impressive

@KSimback While the topline ARR is impressive, what's more eye catching is the dramatic expansion of inference margins from 38% to 70%!