Sabitlenmiş Tweet
Don’t underestimate the difficulty of being able to “verify elsewhere” the untruths that LLMs create.
English
Dr John Flackett
6.8K posts
Dr John Flackett
@drjohnflackett
AI Educator, Speaker & AdvisorㅣPhD Artificial Intelligence | Co-founder & Head of @_AiLab & koolthㅣHonorary Doctor of Science
UK; Kaurna land; Global Katılım Şubat 2012
588 Takip Edilen1.3K Takipçiler
@DynamicWebPaige You will ❤️it so much. Strangely beautiful but deadly at the same time. Let us know how it goes.
English
👋 Greetings from Darwin, about to head into the outback (via bus), wish me luck!
Things I wasn't expecting:
- mysterious-looking WWII oil storage tunnels, which never actually stored oil
- birbs that look vaguely like cassowaries just walking around (actually just orange-footed scrubfowl)
- country western stores with moderately legit cowboy hats and boots
- lots of beautiful national and city parks to walk around in
- zero Starbucks locations in the city center, mostly just lovely small coffeeshops
English
Dr John Flackett retweetledi
I hate the trend of people using AI and then placing the burden of verifying truth and correctness on others.
English
We won’t get to AGI by scaling alone. We’ll need new AI algorithms & approaches to make the next big leap in capabilities. In the end, predicting when/if we’ll get to AGI is useless. Why? Because no one really knows. We’re all guessing!
English
For the past 30 years I’ve been saying that machines that match human intelligence are between 50-100 years away. Today, I say that timeline is still 50-100 years away. Why? Because there’s still so much to figure out. 2/3
English
For the past 10 years, big tech have been telling everyone that AGI (Artificial General Intelligence) is less than 10 years away. The current message is that AGI is less than 10 years away. See the game? 1/3
English
Trust this Gov? What about the next one? We must push back with everything we have to stop this. Our privacy is being eroded & once gone you won’t get it back. 2/2
English
“why would I object to "the state" issuing me with digital proof of ID?”
Because you can either opt out of some of these things & others are siloed. “The state” knowing everything you do WILL, at SOME POINT lead to surveillance & targeting of specific groups. 1/2
Paul Mason@paulmasonnews
I just used my fingerprint to log into my Apple computer; I used facial recognition to pay for an e-bike on my Apple phone; I ordered a repeat prescription on the NHS app that holds every detail of every illness I have ever been treated for on a central server...1/
English
October 2 in London: Alan Kay, Alan Turing, and yours truly:
Tickets (link below) won’t last long.
English
@GaryMarcus See you there @GaryMarcus. Really looking forward to hearing how the discussions go.
English
it's amazing how some people can make a simple agent loop sound so complicated
Ashpreet Bedi@ashpreetbedi
Grifters like this are wasting your time and their Dunning-Kruger opinions should be ignored by serious builders. You either build on a framework or live long enough to roll your own (which is fine btw). Here’s why: 1. The "LLM API in a while loop" is your underlying agentic step - the unit of work in your agent system. You’ll wrap this in a function and call it when you want to run your agent. This is just the starting line, and also where these idiots are usually stuck. 2. As you start building Agents, you turn this function into a class and add logic for prompting, message management, tool calling, retries and error handling. Congrats, you’ve started to roll your own framework. 3. Then, you’ll want to try different models or chain agents to build a workflow. Each has its quirks, custom settings, different response formats, prompting hacks (eg: claude needs "Do not reflect on the quality of the returned search results in your response"). You start adding more layers to your homegrown framework. Till now this is a solvable problem and you can vibecode it. 4. Then things really start to get frustrating. You learn quickly that you need to maintain session history and state across runs because, unlike these grifters, you’re not actually going to run a jupyter notebook in your mom’s basement. Guess what — your agents need a database. You’ll design tables like agent_sessions, wire up session IDs, store/retrieve history and state on each run. Weeks later, you’ll find your schema is inefficient because you forgot to add the right indexes, and now you’ve got to learn about database migrations. FUCCKKKK! Wasn’t this supposed to just be a while loop? And we haven’t even started with RAG, chunking, embedding, retrieval, context management, tool management, MCP, monitoring, and logging. Why did the grifter grift? Now we're weeks into a vibecoded nightmare and the CEO is asking when can we launch. 5. Finally, you confront the real problem: how do I serve this as an API and build a product on top of? You hack together a FastAPI app, throw it in a container, and think you’re done. You breathe a sigh of relief and hand it to your CEO. He hammers it like a madman, chunks start dropping, agents mix context, and you suddenly need to learn about SSE. You implement SSE, things work well for a while and then your requests start timing out and your container memory blows up - memory leaaakk. You blame python, but deep down, you know, you know. You start searching for solutions and come across this code snippet from @AgnoAgi In 25 lines of code you get: ✅ An Agent with memory, state, knowledge, and everything you could ask for. ✅ Access to 1000s of tools and MCP servers. ✅ A production-grade FastAPI app with pre-built SSE-compatible endpoints. You also get a vibrant community and a team of experts to help you every step of the way. This is a systems engineering problem and you get the right tools for the job. You start seeing the light, give up your masochism, and repeat the mantra: JUST USE THE FUCKING FRAMEWORK!
English
OMG 😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂 my belly hurts from laughing too much 😂😂😂😂😂😂 make it stop😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂JFC that donkey😂😂😂😂😂😂😂the chicken😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂
Hakan Kapucu@1hakankapucu
This woman gives a chicken toy to this donkey. He loves it, brays with utters of joy, & funnily plays with it. No innocent being stays indifferent to love.
English
Dr John Flackett retweetledi
Join my new #WomeninTechPowerNetwork, sign up to the waitlist, first 50 subscribers get introductory rate of £29pm, doors open 11am UK time next Monday 1st September:
linkedin.com/posts/sueblack…
English
Dr John Flackett retweetledi
Our @drjohnflackett recently wrote an article for SO: Cultured in @dailyecho & @SotonForward about AI in the arts & culture sector & his 30yr Artificial Intelligence journey, including his PhD in AI & recent Honorary Doctor of Science award for contributions to AI & education.
Daily Echo@dailyecho
Creating environments where people can learn about AI ift.tt/ZaGWKeX
English
PhD application sent to my supervisor 🤞🏼🤞🏼🤞🏼🤞🏼
Totally surprised 😳 at how much I’ve done in terms of research incl a few peer reviewed articles 😅
Wish me luck - I have yet to be accepted 🙏🏼
English
Dr John Flackett retweetledi
We are so sad to share that Dame Stephanie “Steve” Shirley CH has passed away aged 91. Child refugee turned tech pioneer, philanthropist & autism advocate, she founded an all woman software company & changed countless lives. Her light will never fade.
Steveshirley.com
English
@DameWendyDBE @DameStephanie_ Well said, and what an incredible legacy!
English
@drjohnflackett @DameStephanie_ You are too kind, John. I’m very very sad today but there are so many things we have to remember her by. We must celebrate her legacy as much as we mourn her passing
English
Dr John Flackett retweetledi
So sad to hear the news that my great friend and mentor @DameStephanie_ passed away at the weekend. This is the last picture I have of us together when we celebrated her 90th birthday. She was an inspiration to so many people. RIP my dear steve, you changed the world
English