James👁🐝Ⓜ️

Hey there, Mr. Blue We're so pleased to be with you Look around, see what you do Everybody smiles at you.

[James👁🐝Ⓜ️ @jamestaulbut]

Oh, Mr. Blue Sky, please tell us why
You had to hide away for so long (so long)
Where did we go wrong?
Hey there, Mr. Blue
We're so pleased to be with you
Look around, see what you do
Everybody smiles at you. https://t.co/CeFx3N8D5a

[James👁🐝Ⓜ️ @jamestaulbut]

@carwowuk No!

[James👁🐝Ⓜ️ @jamestaulbut]

@MarioNawfal oh my god not a cat!

[James👁🐝Ⓜ️ @jamestaulbut]

@CARmagazine @Jaguar No it’s not a cat!

[James👁🐝Ⓜ️ @jamestaulbut]

@SawyerMerritt No!

[James👁🐝Ⓜ️ @jamestaulbut]

@Jaguar Omg! A jaguar is an agile wild cat. Not a coffin on wheels!

[James👁🐝Ⓜ️ @jamestaulbut]

@SW_Help awful train journey home. Absolutely rammed, this can’t be safe! WAT to PMH 16:30

[James👁🐝Ⓜ️ @jamestaulbut]

Air ambulance lands on Alverstoke village green. I hope whoever needs this is okay, thoughts and prays x

[James👁🐝Ⓜ️ @jamestaulbut]

Consciousness really…“New AI Claude 3 shows signs of Metacognition — A New Era for Humanity & The Science of…” by Pranath Fernando | AI Consultant
https://t.co/SPNG7zm2CZ

[James👁🐝Ⓜ️ @jamestaulbut]

What’s happening @GosportPolice ?

[James👁🐝Ⓜ️ @jamestaulbut]

Helicopter hovering close to Alverstoke Junior School. What’s going on?

[James👁🐝Ⓜ️ @jamestaulbut]

Well here we go… Madonna at the O2 London…

[armand @armand_ruiz]

Training LLMs from scratch costs millions. That's why mostly only Big Tech companies or very well-funded startups can do it today. But do you know why? Let me walk you through it.

𝟭/ 𝗗𝗮𝘁𝗮

In the initial phase of pre-training, you need to curate TB of data and then spend a lot of time pre-processing. This process involves collecting, cleaning, and organizing massive amounts of data to ensure the model trains on relevant and high-quality information. This data collection and preparation is a resource-intensive task, requiring significant time and manpower.

𝟮/ 𝗔𝗜 𝗧𝗮𝗹𝗲𝗻𝘁 𝗮𝗻𝗱 𝗦𝗸𝗶𝗹𝗹𝘀

Developing LLMs requires specialized skills, with top researchers at companies like OpenAI rumored to receive up to $10 million in compensation due to the competitive nature of the field. A team of machine learning, data science, and linguistic experts is essential. They design and refine neural networks, manage training processes, and assess performance. The significant cost of hiring and retaining this skilled workforce is crucial for enhancing the efficiency and accuracy of the model, a vital aspect of AI project success.

𝟯/ 𝗔𝗜 𝗦𝘂𝗽𝗲𝗿𝗰𝗼𝗺𝗽𝘂𝘁𝗲𝗿

Today's AI models, requiring extensive data for high performance, demand significant computing power and GPUs, leading to a focus on AI supercomputers. IBM's Vela AI supercomputer, located in the Washington D.C. data center, exemplifies this with efficient resource allocation, minimal overhead, and powerful hardware like Nvidia A100 GPUs and fast Ethernet links. Developing models like Granite.13b on such supercomputers is costly; for instance, Granite.13b used 256 GPUs for over 1000 hours, with additional training adding 1152 hours, highlighting the substantial financial investment required in advanced AI development.

What does this mean for most businesses?

You don't need to engage in the complex process of training LLMs. If you choose to do so, it must be a very strategic move and can be quite worthwhile, but you need to have a solid ROI plan in place.

Here's my recommendation:

1. Start with LLMs that are already available in the market.

2. Learn how to effectively prompt them for your specific use case.

3. Enhance them with your own enterprise data, using techniques like RAG or fine-tuning.

4. Choose your AI provider wisely, one that will grant you full ownership of the IP and the AI models you tune.

5. Consider optimizations and smaller models to save costs when moving into production.

[James👁🐝Ⓜ️ @jamestaulbut]

Awesome autumnal sunrise this morning as I travel up to IBM London York Rd from my home town, to lead my team. So many colours and so dramatic. ⁦@IBMwatsonx⁩ ⁦@IBM⁩ ⁦@SpinnakerTower⁩

[James👁🐝Ⓜ️ @jamestaulbut]

@MarkHamill @MarkHamill How about a txeets or txits?

[James👁🐝Ⓜ️ @jamestaulbut]

Jaguar driven by AI takes me for a spin through London …not level 5 by any means!

https://t.co/bUcQE920f9

[IBM @IBM]

What do businesses need to know before implementing #AI? We’ll let IBM Director of Research Darío Gil take it from here...

🔒 Protect your data
🌐 Embrace transparency
💡 Implement ethical AI
🏆 Take control

[James👁🐝Ⓜ️ @jamestaulbut]

Would be interested to understand how much energy this model uses…#responsiblecomputing #Sustainability

[Kris Kashtanova @icreatelife]

Google's new flagship AI model, "Gemini," is set to be a direct competitor to GPT-4 and boasts computing power 5 times that of GPT-4.

Trained on Google's TPUv5 chips, it's capable of simultaneous operations with a massive 16,384 chips. The dataset used for training this model is around 65 trillion tokens, and it's multi-modal, accepting text, video, audio, and pictures. Moreover, it can produce both text and images. The training also included content from YouTube and used advanced training techniques similar to "AlphaGo-type" methods.

Google plans to release the Gemini model to the public in December 2023.

[James👁🐝Ⓜ️ @jamestaulbut]

A Hampshire man reports BBC star Chris Packham to police for sniffing goshawk chick - The Times and The Sunday Times. Unbelievable! https://t.co/qwuNN6Bp7J

[IBM @IBM]

It’s time for the #USOpen! 🎾

Explore how the USTA used AI built with #watsonx to bring Match Insights to the US Open app, along with #AI generated commentary for every singles match of the tournament: https://t.co/9Zi695j4Iu..