[alancalverd]

When AI can do all necessary tasks that humans engineers do, they can help solve the problems that they've created in the first place

This reducing AI to the contemptible level of cosmetics and secondary pharmaceuticals.

[hamdani yusuf (OP)]

When AI can do all necessary tasks that humans engineers do, they can help solve the problems that they've created in the first place

This reducing AI to the contemptible level of cosmetics and secondary pharmaceuticals.

The emergence of AI in human society is analogous to the emergence of brain in multicellular organisms. They are predicting machines to help in decision making process.

[alancalverd]

More like a parasite than a brain.

[hamdani yusuf (OP)]

More like a parasite than a brain.

Malfunctioning brains indeed behave like parasites. But the same thing can also be said to other cell types.

[hamdani yusuf (OP)]

"AIs will all become the same" - Platonic Representation Hypothesis Explained

What does Plato have to do with AI? Today we will take a fascinating dive into the world of representation spaces.

[hamdani yusuf (OP)]

?Everything is Going to Be Robotic? Nvidia Promises, as AI Gets More Real

A host of new AI robotic demos, led by Nvidia, have people questioning the future of employment ... and ... reality, while Jensen Huang states that his ultimate ambition is to automate Nvidia entirely. Work has already begun on this, but I show, with interviews and papers, that things might not be so simple to predict.

[hamdani yusuf (OP)]

Ex-OpenAI Employee Just Revealed it ALL!
Here are some important comments.

?You don?t need to automate everything. You only need to automate AI research.?

Surely the fact that they are using the ai to improve its own algorithms will be a major factor in exponential growth

The turning point is self-improvement. When the AI starts to improve itself, then the world as we now know is ended. That's not a bad thing, it's just a statement.

[alancalverd]

So let's implant a gadget which allows my brain to talk to yours. Now I convince you that the next person you meet is an evil persecutor of the righteous, and your sacred duty is to kill him.

Who is liable for murder? How can you prove it? 

[hamdani yusuf (OP)]

So let's implant a gadget which allows my brain to talk to yours. Now I convince you that the next person you meet is an evil persecutor of the righteous, and your sacred duty is to kill him.

Who is liable for murder? How can you prove it? 

Why should I agree with your proposal?
How would you convince me?

Retaliation, reward and punishment, fairness and proportionality are shorthands for how we should react to others' actions and behaviors that generally produce desired effects on the society. But every general rule comes with exceptions.

[alancalverd]

Why should I agree with your proposal?
How would you convince me?

Ask anyone who grooms suicide bombers or Hamas terrorists.

[hamdani yusuf (OP)]

Why should I agree with your proposal?
How would you convince me?

Ask anyone who grooms suicide bombers or Hamas terrorists.

They offered some kind of terminal goal which was regarded as more valuable or important than their own lives. What will you offer?

[alancalverd]

I have a personal terminal goal (a quiet, comfortable suicide to end a happy life in which I made other people happy) which has nothing to do with anyone else. I would offer the same to anyone who asked, and advise them not to obey the orders of parasites.

[hamdani yusuf (OP)]

A terminal goal is a meme. Just like any other memes, they will inevitably compete with one another for the memory space of future conscious entities.
Those memes which promote the survival and sustainability of future consciousness entities will have better chance to survive themselves.

[hamdani yusuf (OP)]

Integrated AI - The sky is quickening (mid-2024 AI retrospective) - LifeArchitect.ai

Read the paper: https://lifearchitect.ai/the-sky-is-q...
The Memo: https://lifearchitect.ai/memo/

Sources: See the paper above.

0:00 Start!
04:17 Large language models
07:18 Datasets
11:11 Synthetic data
13:17 Billboard chart for language models
14:47 Wrapping our mental and emotional arms around AI in 2024
16:43 Countdown to AGI as an average human
17:44 AI is now much smarter than we think

Dr Alan D. Thompson is a world expert in artificial intelligence (AI), specialising in the augmentation of human intelligence, and advancing the evolution of ?integrated AI?. Alan?s applied AI research and visualisations are featured across major international media, including citations in the University of Oxford?s debate on AI Ethics in December 2021.
https://lifearchitect.ai/

[hamdani yusuf (OP)]

Meet Claude 3.5 Sonnet: First Impression of a model Superior to GPT-4o

00:00 Introducing the Cloud 3.5 Sonnet: A GPT-4 Killer?
00:48 Benchmark Performance: Cloud 3.5 Sonnet vs. GPT-4
01:46 Cost and Availability of Cloud 3.5 Sonnet
02:59 Artifacts: A New Way to Interact with Cloud
05:24 Hands-On Demo: Creating a Flappy Bird Game
07:12 Image Understanding Capabilities
08:47 Conclusion and Future Prospects

[hamdani yusuf (OP)]

Claude 3.5 - A Punch to Open AI Where it Hurts.

In this episode, we explore the latest significant development in AI language models - the launch of Claude 3.5 Sonnet by Anthropic AI. This new model not only surpasses its predecessors and competitors but also exhibits exceptional capabilities in complex tasks, coding, and multimodal applications. We delve deep into its features, performance benchmarks, and community reactions. We also discuss its implications for the future of AI and how it stacks up against the current leader, GPT-4 Omni by OpenAI. Join us to discover the potential of Claude 3.5 Sonnet and its role in shaping the next generation of AI.


Timestamps
00:00 Introduction: A Major Shift in AI
00:17 Introducing Claude 3.5 Sonnet
00:52 Claude 3.5 Sonnet vs. GPT-4 Omni
01:40 Features and Capabilities of Claude 3.5 Sonnet
04:36 Vision and Multimodal Capabilities
07:22 Hands-On with Claude 3.5 Sonnet
20:42 Community Reactions and Creations
23:40 Other AI News and Final Thoughts

I just got access to Claude 3.5 Sonnet and tested it for a while. I asked some questions about various things. It generally can answer well about general knowledge which can be found on textbooks. But it often gave wrong answers when the questions aren't widely known or addressed in standard  textbooks. Other AI models I compared with also gave similar results.

[hamdani yusuf (OP)]

Detecting Cancer From a Drop of Blood (The Anti-Theranos)

Detecting cancer from a drop of blood sounds like science fiction, but it may just be around the corner. In this episode, we explore how multiple teams of scientists have used AI to detect cancer.

This news affect me personally and emotionally since both my parents died from cancer.

[hamdani yusuf (OP)]

Open AI's Q* Is BACK! - Was AGI Just Solved?

00:12 New Q* Paper
01:27 Q*
03:09 New Paper
05:17 AlphaGo Explained
08:35 Alphago Search
10:59 Alphacode 2+ Search
14:24 Noam brown On MCTS
17:59 Sam altman Hints at search
19:15 New AGI Approach
20:01 AGI Benchmark
22:20 AGI Benchmark Solved?
24:40 Limits
29:05 Predictions for Future

[alancalverd]

Blood tests for cancer are not new. AI may speed up the interpretation of the test, but the basic chemical pathology is well established.

A rapidly emerging problem for the UK National Health Service is the establishment of new "Community Diagnostic Centres". Good news for me, designing and testing MRI, CT, and other diagnostic suites every other day, but there is a chronic shortage of radiographers  to operate them (it takes 4 years to train) , and once we (i.e. the overworked radiologists, with or without AI) have reached a diagnosis we have no more operating theaters, nor surgeons and anaesthetists (7 years)......So lots of dissatisfied patients.

[hamdani yusuf (OP)]

Recent breakthroughs in AI: A brief overview | Aravind Srinivas and Lex Fridman

GUEST BIO:
Arvind Srinivas is CEO of Perplexity, a company that aims to revolutionize how we humans find answers to questions on the Internet.

A short summary by Claude AI:

I'll summarize the key points discussed in this video about the development of language models and attention mechanisms:

1. Evolution of attention mechanisms:
   - Soft attention was introduced by Yoshua Bengio and Dimitri Bahdanau.
   - Attention mechanisms proved more efficient than brute force RNN approaches.
   - DeepMind developed pixel RNNs and WaveNet, showing that convolutional models could perform autoregressive modeling with masked convolutions.
   - Google Brain combined attention and convolutional insights to create the Transformer architecture in 2017.

2. Key innovations in the Transformer:
   - Parallel computation instead of sequential backpropagation.
   - Self-attention operator for learning higher-order dependencies.
   - More efficient use of compute resources.

3. Development of large language models:
   - GPT-1: Focused on unsupervised learning and common sense acquisition.
   - BERT: Google's bidirectional model trained on Wikipedia and books.
   - GPT-2: Larger model (1 billion parameters) trained on diverse internet text.
   - GPT-3: Scaled up to 175 billion parameters, trained on 300 billion tokens.

4. Importance of scaling:
   - Increasing model size, dataset size, and token quantity.
   - Focus on data quality and evaluation on reasoning benchmarks.

5. Post-training techniques:
   - Reinforcement Learning from Human Feedback (RLHF) for controllability and behavior.
   - Supervised fine-tuning for specific tasks and product development.

6. Future directions:
   - Exploring more efficient training methods, like Microsoft's SLMs (small language models).
   - Decoupling reasoning from factual knowledge.
   - Potential for open-source models to facilitate experimentation.

7. Challenges and opportunities:
   - Finding the right balance between pre-training and post-training.
   - Developing models that can reason effectively with less reliance on memorization.
   - Potential for bootstrapping reasoning capabilities in smaller models.

The discussion highlights the rapid progress in language model development and the ongoing challenges in creating more efficient and capable AI systems.