David Hood

AI, Data, Tech, Business, and Strategery Geek

[David Hood @42RobotsAI]

DeepSeek's benchmarks are a paper mirage

[David Hood @42RobotsAI]

@AnthropicAI New TYPE of AI model benchmark has Anthropic's models crushing the competition. This is a much better reflection of our usage of LLM's to solve real problems. AI Model World Domination Leaderboard: https://t.co/rTN9ET8kvd

[David Hood @42RobotsAI]

@Dhanush_Nehru Looks fantastic

[David Hood @42RobotsAI]

@tyingshoelaces_ Lots of useful and inconvenient truths in this one

[Edward — systemprompt.io @tyingshoelaces_]

My blog on the lie of #agent #frameworks in #ai is catching a bit of 🔥🔥🔥. I'm more excited about the AI Generated Video reviews 👇... Got some exciting news coming soon... https://t.co/Amav7ptMdr

[David Hood @42RobotsAI]

"There is not a single instance of an LLM that has created a superior version of itself" -- @pmarca

https://t.co/mMpaIELkwG

I'll add that transformers (the T in GPT) are also about the same 500 lines of Python code that they were since Google released their paper in 2017...

SLOW YOUR ROLL TALKING ABOUT AGI/ASI/ALL JOBS/ALL DEV JOBS/ETC COMING IMMINENTLY.

Sorry, not sorry, for yelling. Yes, AI and LLM's provide a ton of new, unique value to the world, but the amount of hand-wavy AI nonsense is giga out of control.

[David Hood @42RobotsAI]

@bindureddy It is going to be less than a year away for the next 10-20 years

[David Hood @42RobotsAI]

@haider1 They are weak sauce

[David Hood @42RobotsAI]

@icreatelife As the owner of a small account, I am completely unbiased when I say I agree

[David Hood @42RobotsAI]

@nickfloats Because Apple is and has been for a very long time giga overrated

[David Hood @42RobotsAI]

@0xIlyy Buttons are fun

[David Hood @42RobotsAI]

@haider1 More likely we will get them and they will only have marginal improvements

[David Hood @42RobotsAI]

Can we please stop with "news" about what OpenAI, Google, etc "might" or "may" release?

There's so much hand-wavy, wishful thinking going on right now.

AI is very cool and provides real world value in many domains, but us nerds need to slow our roll big time.

[David Hood @42RobotsAI]

@ThePrimeagen 70% of your followers are trolls

[David Hood @42RobotsAI]

@Dr_Singularity Not without a different architecture from transformers

[David Hood @42RobotsAI]

Saying that everyone is going to build their own AI agents is like saying in 1975 that everyone is going to build their own computers. Yeah, I guess you could do that, but the agents that a regular Joe builds will be colossally less effective than ones built by software development professionals.

This is another one of those "it sounds nice in theory, but is ridiculous in practice" ideas circulating around big tech.

[David Hood @42RobotsAI]

So many of the use cases for AI/LLM's go beyond chat, but that is the frame that just about the whole world is looking at the technology through. Even with a tech slowdown in LLM's, there's still a lot of value to unlock. AI Agents are an extension on the chat perspective.

[David Hood @42RobotsAI]

LLM's can create new recipes, but not new ingredients. Its outputs are seriously constrained by its inputs. Let's be realistic.

[David Hood @42RobotsAI]

2025 AI Prediction: Companies will still struggle to make reliable AI agents because they will be unable to throw away the LLM-centric approach and, therefore, will not be able to make reliable solutions.

[David Hood @42RobotsAI]

@bindureddy We've solved it for our use cases. The key is to stop building LLM-centric solutions where LLM's use deterministic coding as tools. Instead, flip it around and use LLM's as tools inside deterministic coding only at the places where deterministic coding won't work. Reliability++