• Follow
  • Follow
  • Follow
  • Home
  • On-Stage
  • AI
  • Workplace
  • On-Air
  • Coaching
  • About
  • Events
  • Video Gallery
  • Blog
  • Newsletter
  • Contact
Punching above my weight

Punching above my weight

Jul 20, 2023

Dalek Dream

Dalek Dream

Jul 20, 2023

“Good evening”

“Good evening”

Jul 20, 2023

Interviewing Edward Snowden

Interviewing Edward Snowden

Jul 20, 2023

Christmas Clobber

Christmas Clobber

Jul 20, 2023

« Older Entries
Next Entries »

Entradas recientes

  • AI JUST DID SOMETHING VERY WEIRD
  • PUMP UP THE VOLUME
  • FINDING THE NEXT BIG THING
  • TELLING THE TRUTH ABOUT ARTIFICIAL INTELLIGENCE

Comentarios recientes

No comments to show.

Spencer's social

Words and pictures from the digital realm

What a great chat on-stage @talentarena.official. What a great chat on-stage @talentarena.official. Sir Tim Berners-Lee gave loads of great details about how to write apps that don't hand our data over to big tech companies, and whether LLMs are a path to AGI. Also, the niche question of why you can't use apostrophes in URLs.
Good to be back at Fira Barcelona. Me and Sir Tim Good to be back at Fira Barcelona. Me and Sir Tim Berners-Lee on stage at Talent Arena at 11.30. @talentarena.official
Okay, this is exciting - this April, I’ll be hosti Okay, this is exciting - this April, I’ll be hosting the Tech Arena at LEAP 2026. It’s a massive tech show that blew me away last year, and if you can, I’d love for you to pay it a visit. At last year’s Tech Arena we had robots, drones, augmented reality, remote brain surgery, quantum computers, future fashion, all on just our stage. And there are plenty more besides. To get a ticket discount, use the code GPSP25 for general passes and LEPS25 for priority passes. See you in Riyadh, 13-16 April!
Well, this will be pretty epic. Next week I'll be Well, this will be pretty epic. Next week I'll be on-stage chatting with the inventor of the World Wide Web, Sir Tim Berners-Lee in Barcelona. @talentarena.official is co-located with MWC, at the beautiful Fira Montjuic - can't wait! The event runs from March 2-4, and we'll be on the Visionary Stage on March 3rd at 11.30am. Arrive early to get a seat!
Maximum Joy Incoming! I think this may be the bes Maximum Joy Incoming! I think this may be the best CLAWS song yet. The band was invented by my 9 year-old daughter, and between us we’ve created some pretty banging songs, using varying amounts of AI. I’ve been blown away by how good the results have been, and this one is, as they say, 🔥. If you have kids, play it to them. If they like it - if you like it, even a little, please like and repost - my daughter will be over the moon! GEEK DETAILS I’m now understanding how to work with the AI to get the sounds and voices I want. Hence in 9LYVZ you’ll find different singers doing different sections, a rap, and some v nice bits of production: - The whispers, ad-libs and shouts throughout - The strange scream at 0’20 - The cheeky break just before each chorus - That great slow-down at 1’40 (it might be called a Caesura, or a Breakdown - I’m no expert) Wow. Just wow. Or should I say - meow.
My favourite headline of the day, courtesy of @bbc My favourite headline of the day, courtesy of @bbcnews
I've been struggling with one cold after another, I've been struggling with one cold after another, all attacking my voice. And only just remembered how good steaming is. How does it work? Tell me while I'm under my misty blanket.
Just switched on my PC, opened Chrome and was gree Just switched on my PC, opened Chrome and was greeted with this tab. I've never seen anything like it before. Should I be worried, or is it just an update for some adblocker I installed ages ago?
I want to bring you another 3 minutes of joy! It’s I want to bring you another 3 minutes of joy! It’s another song from my and my daughter’s latest album, and for this one I’m afraid I decided to get cocky. “The band is AI, so why don’t they sing about how being human, having emotions, being wild, is better than cold computer logic?” So here are CLAWS singing about a different kind of AI - Animal Intelligence. And once again, I think it rocks! GEEK DETAILS This song is a real example of how you can really riff with an LLM. I had some ideas for the lyrics, but not a complete song, so I described my ideas and let the AI fill in the gaps. I instructed it to create a song where each verse ended in the line “your intelligence needs to be more…animal”, and asked it to make the previous line rhyme (oh, what vision Spen…). I also described the vibe - this should be sung by girls who are fed up being teased by the cooler crowd. And when the lyrics popped out - I wasn’t happy with them. So I suggested some changes, and went again. And again. Some parts of the song I liked, some I didn’t, so I went back and forth, gradually shaping it into something I was happy with. The same went for the music - this one took many many goes through the AI before I got something I liked. The main problem was, after generating quite a few banging numbers previously, everything it was spitting out this time was starting to sound very similar. I found myself wondering if the AI only had a very limited repertoire of chords and sounds for any style of music. As we discussed in my recent AI Masterclass, I felt I was stumbling around in the dark, trying slightly different prompts in case one came up with the goods. And after 20 attempts, it finally did. Yes, it’s another standard rock anthem, but it’s added little tweaks and ad-libs that really elevate it. Listen for the phasing of the instruments at 1’10 and 2’48. The point at 1’17 where the drums drop out for a second. The big shouts at 2’47. I’m not a musician - never claimed to be - but these little additions really make it sound more human. No. More Animal.
How do the latest "Chain of thought", or "reasonin How do the latest "Chain of thought", or "reasoning" AI models work? A tiny morsel from Spencer Kelly's AI Masterclass.
Have any other Substack users received this? Altho Have any other Substack users received this? Although it's difficult, I'd suggest setting up a few email addresses, and choosing which one to use depending on how important/personal the service is. Anyone got any suggestions on alternative email strategies?
Yes, a rabbit with four ears. What of it? If you w Yes, a rabbit with four ears. What of it? If you want exclusive early access to the recording of Spencer Kelly's AI Masterclass, it'll be in this newsletter, delivered to your inbox first thing tomorrow. Please spread this message far and wide, subscribe and enjoy!
Ever tried to generate a specific image (e.g., "A Ever tried to generate a specific image (e.g., "A red box on a yellow box with a stuffed toy on a chair in the corner”) and the AI gives you the right colours, but on the wrong objects in the wrong place? 🚫 In Spencer Kelly’s AI Masterclass, @_mikepound explains why AI is great at vibes but terrible at instructions. The Problem: AI models (like OpenArt) work on semantic themes, not logic. You ask for a "box on a chair in a corner," and it just mashes the concepts together. It’s why AI is great for prototyping ideas, but nightmare fuel if you need specific storyboards. 🎨 The Solution: Control Nets Mike suggests moving beyond basic text prompts to tools like ComfyUI. Using "Control Nets," you can feed the AI a stick-figure sketch or outline. It forces the model to "colour within the lines," giving you the exact composition you need. No more hallucinations. ✍️🤖 But AI is feeding on itself: 😨 Here is where it gets existential. We are running out of natural training data. So, engineers are now using tools like Control Nets to generate Synthetic Data—mass-producing AI images to train future AI models. Which scares the hell out of me. I’ve seen AI trained on AI that’s been trained on AI trained on AI. And it’s not pretty. The Big Question: Is "Synthetic Data" (AI training AI) the inevitable future of tech, or are we heading toward a feedback loop of digital noise and "Model Collapse"? 📉 Reply with your take! 👇 #AI #GenerativeAI #TechTwitter #ComfyUI #SyntheticData
AI has been around for 70 years - but this is why AI has been around for 70 years - but this is why people are only talking about it now.
Best not ask... Best not ask...
"Prompt Engineering" - it’s not a science, it’s an "Prompt Engineering" - it’s not a science, it’s an art. Get better results from AI by following these ludicrous prompting hacks that actually work. 🤖 1️⃣ The "Self-confidence" Technique: If you say, "You are an expert" the results improve. Why? It forces the AI to look at the bit of its training data that's most relevant. After all, you want it to look at "Doctor's notes" rather than "gardener’s notes" on how to remove a limb. 🩺🥀 2️⃣ Be excruciatingly explicit Don’t just ask for a summary. Tell the model exactly how you want that summary formatted (bullet points, tone, length). The model uses every word you give it to calculate the output, so specificity prevents it from guessing. 3️⃣ AI Inception: Struggling to get the right image from Nano Banana? Use a text AI to write the detailed prompt for the image AI. Let the robots talk to each other! 🤖🤝🤖 TELL ME: Be honest in the comments 👇 When you talk to AI, what are your best prompt hacks? #AI #TechTips #PromptEngineering #SpencerKelly #FutureOfWork #ChatGPT #ArtificialIntelligence #CodingLife
I’m really proud of Spencer Kelly’s AI Masterclass I’m really proud of Spencer Kelly’s AI Masterclass - we discussed in-depth all of the latest topics and issues: Practical demos including: ➡ Use AI in a smart way to manage your inbox. ➡ Properly control AI image generation What are organisations doing wrong when implementing AI: ➡ Going in blind, throwing AI at a random problem ➡ Instead, come up with small concrete examples of how LLMs can help, and test them in a controlled environment first AI’s unfixable, fatal security flaw: ➡ LLMs don’t distinguish between instructions and data ➡ This means there’s no way of stopping someone hiding commands that can persuade an LLM to go rogue How should coders work with AI: ➡ Think of LLMs as pretty good, but inexperienced coders ➡ They can only make small tools and utilities in their own Will AI remain “free”: ➡ Is this a standard disruption model, a land-grab of users, with charges brought in after we’re locked into an ecosystem? Is there something different about Grok that allowed it to create inappropriate images: ➡ All image generators are capable of creating those images ➡ But most have better guard-rails at the front end I’ll drop clips from the session in the coming days. And if you came to the Masterclass or you’re a newsletter subscriber, you’ll get first access to the whole recording, plus an extra Q&A. I talk at loads of events about AI and everything else tech, I interview execs, and also train them on how to do what I do on camera and on-stage. If you’d like to know more, get in touch!
This is the biggest mistake most organisations are This is the biggest mistake most organisations are making in adopting AI, according to my guest Michael Pound. Is he right? What successes have you had implementing AI? I'd love to hear about them. Join the conversation: Have you gone big and expensive, or have you followed his advice? #AI #Technology #Innovation #Podcast #GenerativeAI #MachineLearning #SpencerKelly #FutureTech #Science

© Spencer Kelly. All rights reserved. Website byAbi

Privacy Policy

Cookie Preferences