Received the email from them last night. I wish they had spoken to what the difference really is. BUT, admittedly, I have not used Perplexity yet today. Maybe it will be obvious.
Perplexity!
ChatGPT - This is pretty awesome :)
- Thread starter Rick Buffkin
- Start date
Received the email from them last night. I wish they had spoken to what the difference really is. BUT, admittedly, I have not used Perplexity yet today. Maybe it will be obvious.
Perplexity!
Deep Research is a GIANT upgrade to AI, you can find this feature at perplexity, ChatGPT and google Gemini. It creates an 'agent' that reads your request, asks you questions to improve it's understanding, then it becomes a web searching machine.
Heres some AutoMagic Labs research: ChatGPT - Car Shopper JTBD Scoring
It 'thought' for 6 mins before it put out any info. For this post, I asked it for this list:
Heres some AutoMagic Labs research: ChatGPT - Car Shopper JTBD Scoring
It 'thought' for 6 mins before it put out any info. For this post, I asked it for this list:
Citation Count Summary
- AutomotiveMastermind: 3 citations
- Kelley Blue Book (KBB) Study: 2 citations
- Dealer Inspire (Cars.com): 6 citations
- Dealer.com (Cox Automotive) – Accelerate My Deal: 5 citations
- DealerOn – Personalized Offers & Digital Retailing: 4 citations
- Sincro (Ansira) – AWA Awards Product Reviews: 4 citations
- Capital One Auto Navigator – “Wonder Why You Dread Car Buying?”: 3 citations
- Izmocars – “Decision Fatigue and Your Car Dealership”: 3 citations
- Brian Pasch AWA Awards – Sincro Websites: 2 citations
- Cox Automotive – CDK Global Roadster Digital Retail: 2 citations
- Barry Schwartz – “The Paradox of Choice”: 2 citations
Last edited:
IMO, Claude is amazing.
I’m using it head to head with GPT. Claude is improving WITHOUT “new product” announcements. Claude is in bed with Amazon's AWS.
"Claude AI was founded by Dario Amodei and his sister Daniela Amodei, who were both previously high-ranking researchers at OpenAI, with Dario serving as the Vice President of Research; they left OpenAI to establish Anthropic, the company behind Claude, with a focus on developing AI systems prioritizing safety and alignment with human value"
I’m using it head to head with GPT. Claude is improving WITHOUT “new product” announcements. Claude is in bed with Amazon's AWS.
"Claude AI was founded by Dario Amodei and his sister Daniela Amodei, who were both previously high-ranking researchers at OpenAI, with Dario serving as the Vice President of Research; they left OpenAI to establish Anthropic, the company behind Claude, with a focus on developing AI systems prioritizing safety and alignment with human value"
@Carsten, Look around, its changing at light speed. Here is my Slack message to my dev team to start our week:
I must stress to both of you that the evidence of AI acceleration is every where.
AI powered Coding will empower productivity to make senior coders superhuman. Mastery of AI prompting and “command of the domain” is where the real money making will be.
The automotive retail universe is filled with legacy technology that is easily 20 years old. AML‘s LLM centric platform is exactly at ground zero of this new era. I fully expect that the work of a tiny team of domain experts coupled with bleeding edge AI dev tooling can do the work of 50 cube workers.
Buckle your seat belts. It’s gonna be one hell of a ride.
I agree. But to master the prompt is knowing your tools (programming language). It's like OBD scanners. Being able to look up PE1001 isn't it though. You need to know what else is going on. So yes 100% for Domain knowledge.
I spend about 1-2 hours watching YT for the latest changes in Ai recently with some fall back on coding languages.
Friday/Saturday is 1-2 hours for DealerRefresh weekly video.
I don't use Ai work much due to PII concerns. I have zero access to real data but it would be professionally irresponsible since the Ai calls home and our work machines can't handle even 1.5B models.
I use a macbook pro with the m1 max and 64gb of ram. I can get 20 tokens using something like LM studio. A coding IDE is just lag and can't be bothered so I pay the heavier tiers.
This is Windsurf using sonnet 3.5 from Condeium which is Anthrophic. I chose it over Cursor to test first.
I spend 3-4 hours in this IDE a night. I use Trae, Roo code (cline clone), and chatGPT o1 / 4 as backups.
The code base is pretty large and sonnet 3.5 even after 2-3 days of trying to fix the current issue is still not smart enough to code it's way out of the box even by accident. Very few YT videos will talk about the problems that a mature code base presents to Ai. All the other YT videos that push the hype are just as informative as those drop shipping buying pallets to resell are.
But!! as Joe said, "Buckle your seat belts. It’s gonna be one hell of a ride."
I totally agree!
I spend about 1-2 hours watching YT for the latest changes in Ai recently with some fall back on coding languages.
Friday/Saturday is 1-2 hours for DealerRefresh weekly video.
I don't use Ai work much due to PII concerns. I have zero access to real data but it would be professionally irresponsible since the Ai calls home and our work machines can't handle even 1.5B models.
I use a macbook pro with the m1 max and 64gb of ram. I can get 20 tokens using something like LM studio. A coding IDE is just lag and can't be bothered so I pay the heavier tiers.
This is Windsurf using sonnet 3.5 from Condeium which is Anthrophic. I chose it over Cursor to test first.
I spend 3-4 hours in this IDE a night. I use Trae, Roo code (cline clone), and chatGPT o1 / 4 as backups.
The code base is pretty large and sonnet 3.5 even after 2-3 days of trying to fix the current issue is still not smart enough to code it's way out of the box even by accident. Very few YT videos will talk about the problems that a mature code base presents to Ai. All the other YT videos that push the hype are just as informative as those drop shipping buying pallets to resell are.
But!! as Joe said, "Buckle your seat belts. It’s gonna be one hell of a ride."
I totally agree!
AI-Enhanced Development: Key Gains in Speed, Code Quality, and Productivity
p.s. This data is months old! The gains keep on coming!
Sources:
- Speed
- 55% Faster Task Completion: Developers using AI coding assistants solved tasks up to 55% faster.
- 26% More Tasks Completed: Skilled developers completed 26% more coding tasks on average with AI assistance.
- 38.4% More Frequent Iterations: Accelerated development cycles through 38.4% more frequent code compilations.
- Code Quality
- 53.2% Higher Test Pass Rate: AI-assisted code was 53.2% more likely to pass all unit tests.
- 5% Fewer Review Revisions: Code required 5% less rework during review processes.
- Enhanced Readability & Maintainability: Independent assessments showed 2–4% improvements in overall code quality.
- Productivity
- 13.5% More Code Commits: AI tools boosted weekly code commits by 13.5%.
- Weekly Time Savings: Developers saved at least 2 hours per week, allowing more focus on creative and complex tasks.
- Reallocated Effort: Freed-up time enabled developers to concentrate on system design, collaboration, and innovation.
p.s. This data is months old! The gains keep on coming!
Sources:
- GitHub Copilot productivity and quality studies (2024)
- Large-scale enterprise research on AI assistance (MIT/Princeton/U.Penn 2024)
- Developer surveys: Stack Overflow 2023–2024
- JetBrains 2024 report
- GitHub Octoverse 2024
- Uplevel/CIO report on Copilot impact (2024)
- and Dark Reading analysis (2025)
What is GPT smoking?
UI front end development only? How many files are in that code base - 20?
No, I'm sorry. Ai coding as I keep saying just isn't there yet. I might be able to spin up a sort of working blog in react with passable UI but anything beyond that is just marketing pitch that 98% of YT is pushing on people
Ai is great at boilerplate. Create me a new file to do X. I can't match it even with the built in tools. However, let's take this XenForo app.
It will never show any of those stats as being realistic.
CEOs want to hear the above stats.
CTOs want to believe those stats.
Devlopers are begging the gods of great awesome powerful fortune to be true.
UI front end development only? How many files are in that code base - 20?
No, I'm sorry. Ai coding as I keep saying just isn't there yet. I might be able to spin up a sort of working blog in react with passable UI but anything beyond that is just marketing pitch that 98% of YT is pushing on people
Ai is great at boilerplate. Create me a new file to do X. I can't match it even with the built in tools. However, let's take this XenForo app.
It will never show any of those stats as being realistic.
CEOs want to hear the above stats.
CTOs want to believe those stats.
Devlopers are begging the gods of great awesome powerful fortune to be true.
