Deon

Entropic thoughts of a transplant living in Missouri. Just keeping myself sane as i build ai.

[Deon @nifti]

@VictorTaelin similar concept. Copied the grok build /implement persona design as well as the google persona paper to create a simulation engine

[Deon @nifti]

@imjaredz @sachnk @tryramp @DevinAI Basically map each model’s jagged frontier as a multi-dim area plot. Drop the prompt onto the same space, then pick the one whose shape gives the smallest delta covers what’s needed without a bunch of useless overlap. Profit

[OpenRouter @OpenRouter]

Notably, the budget panel was comparable with Claude Fable 5 in performance.

A panel of Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro, fused together, beat solo GPT-5.5 and solo Opus 4.8 outright.

And it landed within 1% of Fable 5 while costing roughly half the price.

[Deon @nifti]

@OpenRouter Composer, DeepSeek, Grok, Opus and GPT5.5 and you have the Zodiac

[Deon @nifti]

@Leohuynh57 1.4 Billion token spent on it this weekend

[Deon @nifti]

@TriCast_ incredible model. it shortens idle time

[Deon @nifti]

@karpathy @jino_rohit ai is moving so fast there is already technical drift

[Deon @nifti]

@tengyanAI decision fatigue is real but the brain is also a muscle

[Deon @nifti]

One of the things an AI programmer has to do is set aside time every weekend to try 10 new ideas out. My schedule is full but its a necessary evil to stay at the frontier.

[Deon @nifti]

@svpino Just need a lot of adhd, trust issues, and a feature rich roadmap planned in the morning. The more you can anticipate , the further you can extend your programming session.

[Deon @nifti]

Okay 3 cent for flash... this was what Google was supposed to charge for Gemini-3.1-flash-lite

[DeepSeek @deepseek_ai]

API is Available Today!

🔹 Keep base_url, just update model to deepseek-v4-pro or deepseek-v4-flash.
🔹 Supports OpenAI ChatCompletions & Anthropic APIs.
🔹 Both models support 1M context & dual modes (Thinking / Non-Thinking): https://t.co/MUPiwkDI8T

⚠️ Note: deepseek-chat & deepseek-reasoner will be fully retired and inaccessible after Jul 24th, 2026, 15:59 (UTC Time). (Currently routing to deepseek-v4-flash non-thinking/thinking).

6/n

[DeepSeek @deepseek_ai]

API is Available Today!

🔹 Keep base_url, just update model to deepseek-v4-pro or deepseek-v4-flash.
🔹 Supports OpenAI ChatCompletions & Anthropic APIs.
🔹 Both models support 1M context & dual modes (Thinking / Non-Thinking): https://t.co/MUPiwkDI8T

⚠️ Note: deepseek-chat & deepseek-reasoner will be fully retired and inaccessible after Jul 24th, 2026, 15:59 (UTC Time). (Currently routing to deepseek-v4-flash non-thinking/thinking).

6/n

[Deon @nifti]

@brian_armstrong between business and trading, those are the two most competitive video games

[Deon @nifti]

@rdd147 pay to play market

[Deon @nifti]

@shoyer close the session, resume in print mode with a for loop that have an agent checks summary for completion and not update

[Deon @nifti]

@lateinteraction @jeremyphoward the amount of time i have to tell the ai to seed the data and save intermediate step to file for benchmarks and iterative runs…

i think I will force it to run jupyter notebooks now. Thanks!

[Deon @nifti]

often times the first solution claude gives is not the best and leads to fragile results. challenge every suggestion.

[Wise @trikcode]

i changed all our "loading..." states to "thinking.."

we are an agentic Al startup now

[Deon @nifti]

Yes anyone can make a MVP app. Now what about the other 20 steps afterwards. I still spend so much time architecting the right production design and requirements. Would be easy if I had unlimited money in AWS. Wouldn’t that be nice. Building against each enterprise constraints is what separates hobbyists and builders.