Dmac
167 posts
Dmac
@bydmac
AI Director & Strategist. Building systems for agentic commerce and helping brands become visible to the AI economy. @Revelancyai
New York, NY 가입일 Ekim 2025
138 팔로잉15 팔로워
Great alpha here. But Perplexity and ChatGPT only overlap on 11% of cited domains. Optimizing for one can leave you blind on the others. You need a cross platform strategy.
nat@natjin
the smartest aeo/geo companies are using perplexity’s search api like a super-weapon they realize you can actually see what is getting ranked and retrieved by LLMs at mass-scale (200m+ daily perplexity queries) vs prompting llms to try to get answers this is real alpha
English
Huge 😯
Glenn Gabe@glenngabe
Focus on ecommerce? Heads-up -> New from Google: AI shopping gets simpler with Universal Commerce Protocol updates UCP onboarding is coming to Merchant Center (in the coming months) -> "We are actively working to onboard more retailers of all sizes to agentic experiences on Google with a simplified UCP onboarding process in Google’s Merchant Center, rolling out over the coming months." More about the latest changes: *UCP can help make online shopping more intuitive and convenient, thanks to a new Cart option that will let agents save or add multiple items to a shopping cart at once from a single store — just as a shopper typically would. *UCP adopters will be able to access a new Catalog capability that lets agents retrieve select real-time product details from a retailer’s catalog where necessary — like variants, inventory and pricing. *Building on existing standards, UCP will also support Identity Linking. That allows shoppers on UCP-integrated platforms to receive the same loyalty or member benefits they would on a retailer’s site when they’re logged in — like pricing or free shipping — making shopping more connected across the web. blog.google/products-and-p…
English
the smartest aeo/geo companies are using perplexity’s search api like a super-weapon
they realize you can actually see what is getting ranked and retrieved by LLMs at mass-scale (200m+ daily perplexity queries) vs prompting llms to try to get answers
this is real alpha
English
We just crossed $5M ARR fully bootstrapped
Introducing CrowdReply 2.0
The new benchmark of ranking in AI Answers
English
@simplifyinAI Not crazy about some of the methodology and scoring process personally, but overall really awesome tool and crazy that it was open-sourced!
English
🚨 BREAKING: Someone just open-sourced a tool that optimizes your website for AI search engines.
It’s called geo-seo-claude. It optimizes any website for AI search engines like ChatGPT, Perplexity, and Claude.
→ Runs full GEO audits with parallel subagents
→ Delivers 60-second visibility snapshots
→ Analyzes structured schema markup for LLMs
→ Exports complete PDF reports
100% Open-Source.
English
Dmac 리트윗함
@daniel_mac8 I remember watching the 2027 report and scared shit out of me… can’t believe we’re here already
English
if you're doing serious research, run perplexity deep research AND gpt5.2-pro side by side
different methodology, different blind spots... together they cover ground nothing else touches (use opus 4.6 to merge both outputs)
gpt takes about an hour to come up with a report tho
English
Update: People are actually seeing ChatGPT ads now.
Expedia ads showing up after single prompts.
The $200K beta is over.
The real experiment begins.
Hot prediction: Gemini ads within 1 month 👀
English
dear grok,
please show this to the best up-and-coming ai creators, filmmakers, and creative minds. i need to give them seedance access.
English
@thedorbrothers @LoganPaul wow! super impressed by your guys turnover time too.
English
We made a $300,000,000 movie starring @LoganPaul with AI in less than 7 days.
Yes, this is 100% AI.
English
My @openclaw agents make nearly no mistakes now. I kinda cracked something.
How?
I employed Chain of Thought (CoT) into every task. Kinda like Ralph, with Claude Code, but better.
Heres how you can implement it for your agents 👇️
The problem right now is context gets lost between prompts/sessions.
Either your window falls off and doesn't grab enough data or your prompt is too big.
One simple fix is to chunk tasks into small bites. We all know this though.
Ralph made this popular with Claude Code to chunk tasks into smaller, more manageable ones.
But, it seems that carrying context and thesis, is like making a copy of a copy. Eventually the original goal is too distorted and you get hallucinations.
Studying how LLMs work under the hood, I may have found a partial solution.
I dubbed it the Cascading Protocol, which to my surprise, is essentially Chain of Thought in LLM training.
How it works, is each sub task is divided into sub tasks first (nothing new here).
From there, each sub task has an expected input, and an expected output.
Almost like a Graph QL interface requiring data to be shared in a preditable way between APIs.
Heres where the magic happens.
Task [Current +1] is not allowed to start until it checks task 1 for completion of its required output.
A typical workflow might look like:
Task 1 starts => Task 1 runs, and finishes => Task 1 self tests => Task 1 validates output and signals the manager to start Task 2 => Manager has Task 2 validate the output of Task 1 => If confirmed; continue. If invalid, send back to manager and/or Task 1 with instructions on what the required input is.
Its far more resource heavy, but it improved outputs significantly. I'd personally rather wait longer for the right answer than have 5 bad answers come back quickly.
I've been testing this with smaller, less capable models, recently and am finding this to be far cheaper to run while maintaining accuracy. Though more testing is necessary.
English
Bad:
"Hire me for $20k per month and I will implement AI within your business."
Good:
"You're spending $500k/month on ads. 20% of leads never get contacted. 15% are out mid-funnel and I know your bottlenecks. Thats $150k a month I can recover for $30k a month."
English