Mildly Magical (@MildlyMagical) header
Mildly Magical (@MildlyMagical) profile picture

Original thought, critical thinking, problem solving, common sense, loyalty and respect. Trust is the Moat.

Joined August 2023
Mildly Magical retweeted
show this video to anybody that is an old school programmer that hates AI this is one of the best videos I’ve seen all month.
1,138
6,580
1,361
47,665
4,403,877
A new trend I see are these engagement farming accounts that put devs using AI against people writing their code by hand. Don't fall for it.
1
3
71
Mildly Magical retweeted
BREAKING: Tesla’s new Roadster just flew, pulling a barrel roll over a Tesla Semi. @elonmusk: "I've seen a lot of things, but I've never seen that before." "We call that the kickflip."
223
774
249
7,854
1,686,944
Wow
NEW: Tesla Roadster does a mid air summersault from a standstill. WHAT 🤯🤯🤯
2
67
We are making a living on old school developers writing code on typewriters thinking production AI doesn't exist. The last dev I met with, just this week, had the same attitude, as I was sitting there taking his customer and there was nothing they could do about it. We are not dumb enough to say what our products are on the scorned dev threads. @LeoBuilds_
2
76
sweet
Grok Bot now has its own email. Bot can use it to sign up for services, contact businesses for you, or schedule time with someone.
2
2
124
New phrase that seem to produce noticeably better results, for me. "...enable us to move fast without sacrificing thoroughness. I want both"
46
Replying to @LeoBuilds_
@LeoBuilds_ @IndependentEco I gave my product an AI advisory board (8 seats, including a Red Team) that can't touch the code. It runs as its own Claude Code session, walled off from the session that engineers the product. The only bridge between them is me. The board • its own folder and charter • read-only snapshots of the repo, issues and customer records • no shell, every MCP tool denied, writes only to memos/ and outbox/ • never sees the build session's live state or priorities, on purpose: I want a view formed from the code and the customer, not an echo of the build agent • output: a one-page card with recommendation, confidence, evidence, dissent, and what would change its mind The walls live in settings deny rules, not in the prompt. The glue is a Claude Code mod (plugin hooks), board-link: • /ask-board <q> in the build session writes a dated question to the board's inbox • the board answers headless (claude -p, Opus 5.5, max effort) under its own walls • a toast when its memo lands • /board lists every card with its decision line; Read opens it in full • Send to desk drops it into my prompt ending "My decision: ". I add the ruling and press Enter A card counts as relayed only when it was in a prompt I actually sent. The one wall I didn't lift: the board never messages the build session directly. A peer agent's message is data, not a ruling. Advice becomes a decision only when a human presses Enter.
3
1
129
Even the OpenAI team uses Claude 🤷‍♂️🤣
Replying to @b_nnett
that and canceling claude
1
4
81
The new Claude Mods are really cool! I am kind of geeked out on them.
2
50
Mildly Magical retweeted
David Grusch is saying the CIA can legally silence you if you knowingly reveal their secret UFO technology. That’s what Presidential Findings are used for.
64
161
10
1,309
65,457
I have found, empirically, this is very true.
Paper shows multi-agent teams waste over two-thirds of actions on coordination failures: stale messages, role confusion, and lost shared plans. For solo: limit AI helpers to few specialized roles with clear prompts. Maintain one persistent shared plan. Prioritize only high-impact steps. Add frequent checkpoints on long tasks to prevent drift. Scale adds cost, not gains.
1
45
🤣 I'm dead 🤣
AI is getting so crazy 😂
1
1
48
I hope you are not being scammed by OpenAI. $3000 in usage in 15 minutes with 2 agents running. I watched this happen live. Some poor dev is being drained of his funds while trying to ship his app. This is just wild!!
One GPT 6 Astra agent on Ultrafast drained my entire $500 plan for the week in 15 minutes. Then it burned through 95% of the $2,500 in credits OpenAI gave me for cutting my usage in half. 3,103 credits left out of 62,500. OpenAI is terrible.
38
THE King
Facts check out. Jesus is King.
1
1
23
Awesome
Yep, we've known. Christ is King.
16
Stunting on them 🤣
Best way to sandbox an AI is to put it on a Delta flight – it will have no chance of accessing the Internet!
20
Seeing the pattern?
Cancelled every ChatGPT subscription I had. They quietly halved usage yesterday, then dressed it up as "efficiency gains." It's not. GPT-6.1 Sol is the same GPT-6 Sol underneath, just RL-tuned to look better on paper. Keep paying and every lab follows this playbook cutting usage is just good margin for them. How long until every lab tries the same move?
1
60
Some think I'm just being hard on OAI. I am. It's well placed. It is not hard to understand the facts. There is a reason Elon openly calls out this company as being unethical.
I JUST CANCELLED ALL OF MY CHATGPT SUBSCRIPTIONS. OpenAI just scammed everybody and BridgeMind is now starting a boycott against OpenAI. They cut the usage in half yesterday and said better efficiency makes up for it. This is a lie. GPT 6.1 Sol is just GPT 6 Sol with RL to look better on benchmarks. If we keep paying, Anthropic and every other lab will follow. Cutting usage is profitable for them. Do not let OpenAI get away with this.
1
1
71
Mildly Magical retweeted
MLKBench testing equal life value across races, no skin-color skews in tradeoffs or decisions, would match the character-not-color standard. Solid idea for any model eval suite.
1
1
22