Carlos E. Perez @IntuitMachine Twitter profile

Pinned Tweet

Carlos E. Perez

2 months

Quaternion Process Theory In this talk, I expand upon Daniel Kahneman's foundational Dual Process Theory to introduce my approach, which I call the Quaternion Process Theory. This refined cognitive framework proposes four distinct modes of thinking, enhancing our understanding…

7

46

238

Last Seen Profiles

@Gofblug

@zimZalla

@_tutsy

@beeonaposy

@CirnoFumofumo

@ColtonHowardTV

@microbit_edu

@itskaiwan

@KittyKartRacing

@jandakembangstw

@FastBreakCards

@MargareteRosali

@INABVESV

@Hufonime

@acereduau

@TaiHamilton_

@NotBrunoAgain

@JWEmomotaroband

@stw_pdg

@CarlyFiorina

@JeffSonnenfeld

@dcldal

@KSAmofaFRA

@LaCiteNantes

@ClareChika

@MystiicReal

@GoldenAmateur

@DrNeilHudson

@suhpremacistJ

@KappaSigmaHQ

@MemoriesCreates

@RubinsteinOk

@Ryo_cos76

@MedLabNigeria

@TopBoygym

@rivoxyz

Carlos E. Perez

@IntuitMachine

3 years

The state of machine learning practice:

80

2K

23K

Carlos E. Perez

@IntuitMachine

3 years

The brain's consensus algorithm demonstrated:

134

3K

11K

Carlos E. Perez

@IntuitMachine

3 months

Groq is a Radically Different kind of AI architecture Among the new crop of AI chip startups, Groq stands out with a radically different approach centered around its compiler technology for optimizing a minimalist yet high-performance architecture. Groq's secret sauce is this…

105

732

4K

Carlos E. Perez

@IntuitMachine

1 month

Why worry about global debt when this is happening in AI?!

274

4K

Carlos E. Perez

@IntuitMachine

8 months

LLMs are glorified autocompleters. We can say the same about human vision. Here's a demo of autocomplete mode!

139

426

4K

Carlos E. Perez

@IntuitMachine

1 year

#ChatGPT to video via #stablediffusion and other AI tools.

139

596

3K

Carlos E. Perez

@IntuitMachine

7 months

Introducing StreamingLLM. Imagine chatting with an AI assistant that can contextually reference your conversations from weeks or months ago. Or summarizing reports that span thousands of pages. StreamingLLM makes this possible by enabling language models to smoothly handle…

56

476

3K

Carlos E. Perez

@IntuitMachine

27 days

Game Over for traditional ML methods

96

185

2K

Carlos E. Perez

@IntuitMachine

10 months

Simple #Bard tip, just grab the image of an equation and have it rendered in Latex.

66

299

2K

Carlos E. Perez

@IntuitMachine

6 months

1/n Breaking News! OpenAI has uncovered an emergent new cognitive capability, yet nobody is demanding answers! We are distracted by OpenAI governance politics and not the real issue!!!

70

338

2K

Carlos E. Perez

@IntuitMachine

7 months

Introducing Promptbreeder. Promptbreeder employs large language models like GPT-3 to iteratively improve text prompts. But here's the magic - it doesn't just evolve the prompts themselves. It also evolves how the prompts are generated in the first place. Let's break it down.…

32

269

2K

Carlos E. Perez

@IntuitMachine

7 months

Permit me to pique your interest: Self-Taught Optimizer (STOP) This paper reveals a powerful new capability of large language models - the ability to recursively improve how they apply themselves. The authors show that models like GPT-4 can optimize code that leverages the…

44

320

2K

Carlos E. Perez

@IntuitMachine

6 months

Large language models (LLMs) and knowledge graphs (KGs) are complementary technologies that balance each other's strengths and weaknesses when combined: - LLMs have a strong capability for understanding and generating natural language, but can sometimes hallucinate facts. -…

26

297

1K

Carlos E. Perez

@IntuitMachine

6 months

OpenAI to announce GPT-4 with a 128k context window! Time to revisit everything!!!

52

181

1K

Carlos E. Perez

@IntuitMachine

7 months

Confirmation that AGI is indeed here! The classic argument made over 30 years ago by Fodor and Pylyshyn - that neural networks fundamentally lack the systematic compositional skills of humans due to their statistical nature - has cast a long shadow over neural network research.…

51

318

1K

Carlos E. Perez

@IntuitMachine

2 months

I suspect many AI projects will end up in ruin because its developers are just muddling around getting their dopamine hits from the deluge of micro-events about AI. They focus only on the trees but can't see the forest!

44

306

1K

Carlos E. Perez

@IntuitMachine

10 months

An extremely useful trick for #Bard . Grab a screen capture of any text (i.e., #GPT4 generated table) and convert it to actual text! It's OCR for free.

51

214

1K

Carlos E. Perez

@IntuitMachine

2 years

Written 36 years ago, perhaps one of the most important books on AI ever written. Absolutely relevant even today!!

29

167

1K

Carlos E. Perez

@IntuitMachine

5 years

A coincidence today, I did the same thing as @elonmusk . I took my child to play the piano at the home for the elderly.

Elon Musk

@elonmusk

5 years

Took my son to play piano for the seniors home in Pasadena. It was lovely to see them smile ♥️

1K

3K

108K

20

26

1K

Carlos E. Perez

@IntuitMachine

6 months

An ontology for prompting. Components: - Instructions: Short prompts that guide LLM reasoning format and structure - Rationales: Intermediate reasoning steps generated during CoT - Exemplars: Input-output examples that demonstrate target reasoning pattern - Environments:…

17

183

1K

Carlos E. Perez

@IntuitMachine

3 months

AI's Secret Pattern: The Surprising Role of Fractals in Neural Networks In the realm of artificial intelligence (AI), a groundbreaking discovery has emerged, challenging our conventional understanding of neural network training and optimization. This revelation centers around…

32

254

1K

Carlos E. Perez

@IntuitMachine

1 month

Agentic AI is the next wave!

28

176

1K

Carlos E. Perez

@IntuitMachine

10 days

Breaking News: A purported 1.5B parameter model called GPT-2 chatbot has been released and everyone is stunned!!

43

71

997

Carlos E. Perez

@IntuitMachine

4 months

Sam Altman reveals in an interview with Bill Gates (2 days ago) what's coming up in GPT-4.5 (or GPT-5): On multimodality: Sam predicts that the ability to incorporate speech, images, and video will be an important milestone in the next two years. He mentions that OpenAI has…

27

163

934

Carlos E. Perez

@IntuitMachine

2 years

So mind-boggling that the main discovery in research is a specific incantation (i.e. “Let’s think step by step”). Does anyone not recognize how insane this appears?!

Aran Komatsuzaki

@arankomatsuzaki

2 years

Large Language Models are Zero-Shot Reasoners Simply adding “Let’s think step by step” before each answer increases the accuracy on MultiArith from 17.7% to 78.7% and GSM8K from 10.4% to 40.7% with GPT-3.

59

573

3K

34

120

920

Carlos E. Perez

@IntuitMachine

4 months

26 Prompting Tips 1 - No need to be polite with LLM so there is no need to add phrases like “please”, “if you don’t mind”, “thank you”, “I would like to”, etc., and get straight to the point. 2 - Integrate the intended audience in the prompt, e.g., the audience is an expert…

37

157

903

Carlos E. Perez

@IntuitMachine

5 months

OpenAI just came out with their Prompt Engineering guide:

OpenAI Platform

Explore developer resources, tutorials, API docs, and dynamic examples to get the most out of OpenAI's platform.

platform.openai.com

9

155

895

Carlos E. Perez

@IntuitMachine

5 months

1/n Was December 8th, 2023, the day when we've come to realize that AGI technology has been democratized? That it cannot be confined to the few and the GPU-rich? Let me explain to you what happened yesterday.

25

154

854

Carlos E. Perez

@IntuitMachine

1 year

Some more experimentation with #stablediffusion Driving video ()

41

92

837

Carlos E. Perez

@IntuitMachine

10 months

Oh my gosh! You can import several documents into Claude 2 and ask the relationship between the concept found in each document. It's conceptual blending on steroids! This is insane!

26

154

841

Carlos E. Perez

@IntuitMachine

5 months

1/n Breaking News! Prompt Engineering for the Win! Instruct fine-tuning has been discovered to be unnecessary. Prompting is all you need! A recent research paper provides compelling evidence that the extensive fine-tuning used to "align" large language models into helpful…

25

180

837

Carlos E. Perez

@IntuitMachine

7 months

Introducing SocraticAI. For too long, the capabilities of large language models have been constrained by their reliance on human-crafted prompts. SocraticAI provides a more natural paradigm for AI collaboration and reasoning. SocraticAI simulates fluid human discussion through…

33

143

829

Carlos E. Perez

@IntuitMachine

3 months

Wow! Teachers are now inserting Trojan Horses in their assignments!

Jasper

@0xjasper

3 months

Holy shit tiktok discovered prompt injection

80

509

5K

30

91

812

Carlos E. Perez

@IntuitMachine

2 months

Nvidia's Blackwell isn't taking any prisoners.

10

88

807

Carlos E. Perez

@IntuitMachine

1 year

I don't know what to make about this development. Alpaca is surprisingly very good. The claim here is the training can be done in 5 hours on a single RTX 4090. Have GPT-like models been democratized overnight?!

AK

@_akhaliq

1 year

alpaca-lora: Code for reproducing the Stanford Alpaca InstructLLaMA result on consumer hardware github:

25

487

2K

20

143

769

Carlos E. Perez

@IntuitMachine

4 months

LLMs that are "lying" apparently have a recognizable signature.

Andy Zou

@andyzou_jiaming

7 months

In fact, we find LLMs exhibit different brain activity when they express their true beliefs vs. when they lie (see figure).

13

81

492

15

115

767

Carlos E. Perez

@IntuitMachine

2 years

@kamilkazani Better image:

14

53

724

Carlos E. Perez

@IntuitMachine

6 months

1/n Let me start a thread that speculates what OpenAI's Q* (Q-star) may likely to be. To narrow the scope of our exploration, let's assume that it's a derivation of a Reinforcement Learning approach (i.e., Q-learning) applied to LLMs like GPT. Will Q render judgement on…

22

121

749

Carlos E. Perez

@IntuitMachine

1 year

"OpenAGI, an open-source AGI research platform, specifically designed to offer complex, multi-step tasks and accompanied by task-specific datasets, evaluation metrics, and a diverse range of extensible models"

GitHub - agiresearch/OpenAGI: OpenAGI: When LLM Meets Domain Experts

OpenAGI: When LLM Meets Domain Experts. Contribute to agiresearch/OpenAGI development by creating an account on GitHub.

github.com

5

155

734

Carlos E. Perez

@IntuitMachine

3 months

The Hidden Harmony in AI's Complexity: How Different Algorithms Whisper the Same Truth An exciting discovery revealed in this paper is that very different machine learning algorithms and neural networks can encode surprisingly similar representations of data, even though their…

17

170

728

Carlos E. Perez

@IntuitMachine

3 months

1/n What in the world is Sora's "diffusion transformer model"? A diffusion transformer model is a type of generative model for images, video, and other data that combines transformer architectures with diffusion probabilistic models. Here are some key details: - Diffusion…

9

215

727

Carlos E. Perez

@IntuitMachine

2 months

Some dude in Norway gave LLMs an IQ test and Claude 3 scored 101. (source in ALT)

47

153

712

Carlos E. Perez

@IntuitMachine

3 months

1/n Introducing RAPTOR Existing RAG methods suffer from a major limitation: they can only retrieve short, contiguous passages of text. This restricts their capacity to represent cross-document discourse structure and leverage thematic information scattered across lengthy…

13

128

712

Carlos E. Perez

@IntuitMachine

8 months

Introducing Thought Cloning Thought Cloning could enable a revolutionary leap in AI capabilities. For the first time, agents would not just blindly mimic human behaviors, but gain insight into the underlying thought processes behind those behaviors. Just as language transformed…

8

138

685

Carlos E. Perez

@IntuitMachine

7 months

Let's discuss Step-Back Prompting Step-Back Prompting is like taking a step back to see the bigger picture before diving into the details. It's based on the observation that we humans often simplify complex problems by first identifying the key, high-level concepts. We extract…

17

110

684

Carlos E. Perez

@IntuitMachine

3 months

1/n An ontology of Large Language Model (LLM) powered Multi-Agents - Single LLM-based agents have shown promising capabilities such as planning, tool use, memory, and decision making. This has motivated research into multi-agent systems. - LLM-multi agent (LLM-MA) systems aim to…

9

141

666

Carlos E. Perez

@IntuitMachine

2 years

IMHO, diffusion models are as big a breakthrough as transformer models. It's a rare development when an architecture requires fewer compute resources than previous proposals.

What are Diffusion Models?

[Updated on 2021-09-19: Highly recommend this blog post on score-based generative modeling by Yang Song (author of several key papers in the references)]. [Updated on 2022-08-27: Added classifier-f...

lilianweng.github.io

11

88

611

Carlos E. Perez

@IntuitMachine

1 year

@ben_levinstein Let me check:

8

19

593

Carlos E. Perez

@IntuitMachine

2 months

Claude 3 may have a problem. Its responses may be too intelligent for most humans to comprehend. The problem with too much smarts is that you attract a smaller audience. That doesn't work for our dopamine culture.

92

41

591

Carlos E. Perez

@IntuitMachine

4 months

1/n An ontology for hallucination mitigation techniques in Large Language Models (LLMs). Prompt Engineering category A. Retrieval Augmented Generation (RAG) - Before Generation: Strategies where information retrieval happens before text generation, e.g. LLM-Augmenter -…

17

104

586

Carlos E. Perez

@IntuitMachine

1 year

There's a lot to parse in Geoffrey Hinton's explanation as to why he realized that deep learning systems like GPT-4 are more efficient intuition machines than humans. He formerly believed that we needed to model the brain.

20

139

578

Carlos E. Perez

@IntuitMachine

11 months

Finally a paper that validates one crucial difference between GPT-4 and GPT-3.5. This is the ability to perform self-repair.

9

139

573

Carlos E. Perez

@IntuitMachine

7 days

1/n Math Meets AI: Kolmogorov-Arnold Networks Unleash the Power of Composition Imagine a world where deep learning models, the enigmatic engines driving the AI revolution, are no longer shrouded in mystery. What if we could peer into their inner workings, understand their…

33

283

1K

Carlos E. Perez

@IntuitMachine

7 months

Let's talk about Vec2Text This paper introduces a powerful new technique for inverting text embeddings back to their source texts. The method, Vec2Text, demonstrates for the first time the ability to recover full text sequences from state-of-the-art neural text encoders. Through…

13

93

542

Carlos E. Perez

@IntuitMachine

3 months

1/n No Search, No Problem: Achieving Grandmaster Level Using Only a Transformer A new research paper presents a groundbreaking advancement in chess-playing artificial intelligence, demonstrating for the first time that it is possible to train a neural network to play chess at a…

6

114

545

Carlos E. Perez

@IntuitMachine

7 months

Introducing DoLa Have you ever felt frustrated by the factual inconsistencies and falsehoods generated by large language models? As helpful as these models can be, their tendency to "hallucinate" incorrect information threatens their reliability and hinders real-world…

10

94

539

Carlos E. Perez

@IntuitMachine

7 months

Let's discuss the approach of MEMWALKER At its core, MEMWALKER transforms how AI systems process long-form knowledge - and this paradigm shift unlocks tremendous value. For far too long, language models have been confined by fixed context lengths that severely limit their…

9

80

536

Carlos E. Perez

@IntuitMachine

2 months

Interesting that many have convinced themselves that LLMs can't do logic reasoning. They actually can do much more, they can do semiotic reasoning! According to C.S. Peirce, this is a superset of logic.

Tsarathustra

@tsarnick

2 months

Stephen Wolfram: AI language models are discovering things about the world in the same way that Aristotle did

43

156

811

19

110

535

Carlos E. Perez

@IntuitMachine

5 months

The Observe-Orient-Decide-Act (OODA) Loop and Prompting Patterns.

10

101

524

Carlos E. Perez

@IntuitMachine

3 months

BlackMamba Mixture of Experts BlackMamba is an novel architecture which combines state-space models (SSMs) with mixture of experts (MoE). It uses Mamba as its SSM block and switch transformer as its MoE block base. BlackMamba is extremely low latency for generation and…

5

107

527

Carlos E. Perez

@IntuitMachine

1 month

Breaking! SambaNova releases open-source LLM that demolishes DBRX! Breakneck progress!

36

94

499

Carlos E. Perez

@IntuitMachine

5 months

1/n This Google Gemini demo reveals insightful design and prompting methods for building AI-powered user interfaces. Let me take you down the details. Video from @deepmind

11

95

520

Carlos E. Perez

@IntuitMachine

4 months

Here's Steve Jobs, in 1985, talking about his dream that we one day may have a conversation with (virtual) Aristotle. He also mentions how the computer industry was in the dumps, but has faith that it'll bounce back.

21

94

493

Carlos E. Perez

@IntuitMachine

8 months

Introducing Boolformer Machine learning techniques like deep neural networks have achieved remarkable success on perceptual tasks, yet their performance on logic and reasoning problems remains limited. This presents a major barrier to developing transparent and trustworthy AI.…

5

111

496

Carlos E. Perez

@IntuitMachine

2 months

I'm beginning to suspect that Claude 3 has a previously unidentified kind of emergent cognitive capability that perhaps eludes standard benchmarks. What could this be?

25

43

468

Carlos E. Perez

@IntuitMachine

4 years

Solo papers are going back into style! They are great because they are like ambitious manifestos that describe a unique idea and boldly broadcast that this is my idea and my idea alone! Here are the latest ones from prominent AI researchers.

12

112

461

Carlos E. Perez

@IntuitMachine

5 months

The FunSearch paper by DeepMind that was used to discover new mathematics is an example of searching through generative patterns and employing evolutionary methods to creatively conjure up new solutions. This is a very general principle that lies at the core of creativity.…

13

76

468

Carlos E. Perez

@IntuitMachine

1 month

1/n The unassailable logic of debt and AGI is this: (1) debt is borrowing from the future, (2) If AGI's future is unlimited abundance, then (3) why worry about today's debt when it's obviously going to be paid for in the future!

89

85

467

Carlos E. Perez

@IntuitMachine

7 months

Let's discuss Symbol Tuning in LLMs As AI practitioners, we all want language models that can adapt and learn like humans. But our current LLMs fall short - they rely too heavily on their pre-existing knowledge and hard-coded prompts. This leaves us stuck constantly tweaking…

9

85

463

Carlos E. Perez

@IntuitMachine

5 months

I'm in shock! The numbers don't lie! A year later, after GPT-3.5 was released, you have an open-source LLM (i.e., Mixtral) that can run entirely within 2 80GB GPUs. Is GPT-4 level capability reachable in open source in the next few months? Let's not forget, many freaked out…

Philipp Schmid

@_philschmid

5 months

We just got more details on Mixtral 8x7B from @MistralAI 🧠 Mixtral is sparse mixture of expert models (SMoE) with open weights outperforming existing open LLMs like Meta Llama 70B.🤯 💪🏻 TL;DR: ⬇️

2

57

238

17

75

432

Carlos E. Perez

@IntuitMachine

5 months

Key vulnerabilities of GPT-4: 1. Fine-tuning API can remove or diminish safety guardrails, causing the model to produce harmful outputs or assist with dangerous requests 2. Fine-tuning can make the model generate targeted misinformation against public figures 3. Fine-tuning…

15

89

449

Carlos E. Perez

@IntuitMachine

1 month

Wait! Are most medical doctors really just stochastic parrots or glorified autocompleters?

90

30

448

Carlos E. Perez

@IntuitMachine

3 months

1/n Prompt engineering has emerged as a pivotal discipline enabling more sophisticated and nuanced interactions with AI systems like large language models (LLMs). Prompts serve as the interface through which users communicate desired outcomes to these generative models.…

4

72

438

Carlos E. Perez

@IntuitMachine

3 years

@paulkrugman Live Action Role Play (LARPing). Once the FBI started knocking on their parent's doors, it was all over!

10

29

415

Carlos E. Perez

@IntuitMachine

7 months

Introducing Neural Developmental Programs Nature has an ingenious way of creating complex, functional neural networks through a process of growth and self-organization. This biological strategy stands in stark contrast to how we currently design artificial neural networks - by…

12

85

437

Carlos E. Perez

@IntuitMachine

6 months

Traditional thought prompting methods for large language models (LLMs) face significant limitations that constrain their performance and applicability. XOT offers a compelling solution that definitively addresses these pain points: - Lack of world knowledge - LLMs often struggle…

9

85

424

Carlos E. Perez

@IntuitMachine

3 months

1/n Scientists Invent an "Aha Moment" Generator for AI Systems It's a familiar scene - you've written an essay or solved a math problem, feeling assured you have the right solution. But upon review, it becomes clear your original logic was flawed in places. If only you had the…

4

103

412

Carlos E. Perez

@IntuitMachine

1 year

GPT-4 is a tool for thinking harder. Most people need to be made aware of how to wield its power. Get the missing instruction manual.

22

67

406

Carlos E. Perez

@IntuitMachine

18 days

AlphaLLM: An LLM that Learns and Improves Itself Large Language Models (LLMs) have revolutionized the field of Natural Language Processing, demonstrating remarkable capabilities in various tasks. However, they still struggle with complex reasoning and planning, often requiring…

5

128

449

Carlos E. Perez

@IntuitMachine

1 month

1/n An Ontology for Agentic AI Agentic AI systems are ones that can perceive their environment and take actions to achieve their goals. They act independently, without direct human control, continuously interacting with their surroundings over time. The concept of Agentic AI has…

9

88

409

Carlos E. Perez

@IntuitMachine

1 year

The Voight-Kampff test. Dialog generated by #ChatGPT , video generated by a whole bunch of other AI tools. Like if you want me to render the conclusion!

29

54

403

Carlos E. Perez

@IntuitMachine

3 months

K-Level Reasoning LLMs Imagine the potential of artificial intelligence to not just participate in strategic games but to excel by adopting the thought processes of its opponents. The latest research on Large Language Models (LLMs) has unveiled a groundbreaking methodology known…

11

83

407

Carlos E. Perez

@IntuitMachine

3 months

Wow! Brave browser now has Mixtral!

8

48

401

Carlos E. Perez

@IntuitMachine

1 year

I think too many are thinking of GPT-4 in the wrong way. They are all thinking of how to automate tasks (i.e., think less) rather than thinking of how GPT -4 can allow one to think better and thus harder.

A.I. for Thinking Harder and not Thinking Easier

Steve Jobs has this analogy that computers are bicycles of the mind.

medium.com

28

82

399

Carlos E. Perez

@IntuitMachine

6 months

All my books are now available as GPTs. This is only going to get better and more capable! Books don't hold contextual conversations, GPTs do! Do you already have your wagon or are you just started thinking about it?

15

42

403

Carlos E. Perez

@IntuitMachine

4 months

Here's the skinny: 1 - Design a Domain Specific Language (DSL) 2 - Code it in a symbolic engine 3 - Generate synthetic data 4 - Train your Large Language Model (LLM) 5 - Profit Are there any questions left?

Google DeepMind

@GoogleDeepMind

4 months

AlphaGeometry is a system made up of 2️⃣ parts: 🔵 A neural language model, which can predict useful geometry constructions to solve problems 🔵 A symbolic deduction engine, which uses logical rules to deduce conclusions Both work together to find proofs for complex geometry…

11

85

526

12

73

398

Carlos E. Perez

@IntuitMachine

5 months

LoRAMoE?!

1

67

395

Carlos E. Perez

@IntuitMachine

6 months

1/n It should be pretty obvious now that a 7-14B model can best GPT-4 in specialized domains. This realization torpedoes GPU-rich firms from establishing a monopoly. One can leverage extreme asymmetric information arbitrage in the long-tail of LLM applications.

22

44

389

Carlos E. Perez

@IntuitMachine

7 months

Introducing xVAL - the next evolution in numeric encoding for natural language processing! Language models have achieved impressive results on textual data. But analyzing numbers - a fundamental building block of math, science, and life - remains a glaring blindspot. Existing…

10

69

392

Carlos E. Perez

@IntuitMachine

1 year

234k neurons, and it does this????!

Massimo

@Rainmaker1973

1 year

Bumblebees change their flight patterns differently when they have to pass through a tight space based on their size, indicating that they have some idea of their own size and shape despite their simple nervous systems [video: ]

59

668

4K

28

48

385

Carlos E. Perez

@IntuitMachine

1 month

Where do you think OpenAI is finding its "superhuman" training data to imitate?

82

30

390

Carlos E. Perez

@IntuitMachine

6 months

1/n Evidence that Reasoning arises from Intuition Compelling evidence that both humans and LLMs rely significantly on intuitive, inductive processes rather than pure logical deduction when reasoning. Here are three key reasons why: First, the finding that reasoning accuracy…

9

73

383

Carlos E. Perez

@IntuitMachine

6 months

1/n Retrieval augmented generation (RAG) architectures have become popular for knowledge-intensive NLP tasks. However, they face some primary problems: 1. Imperfect retrieval providing irrelevant or distracting passages along with useful context. 2. Full context augmentation…

4

63

393

Carlos E. Perez

@IntuitMachine

2 months

Why are Claude's responses superior to other LLMs? The field of generative AI has seen remarkable advances in recent years, with the development of powerful language models like the Transformer. These models operate primarily through abductive reasoning, generating plausible…

18

76

390

Carlos E. Perez

@IntuitMachine

5 months

Is prompt engineering dead? No, it's SoTA. OpenAI achieves 98% accuracy via prompt engineering!

21

54

388

Carlos E. Perez

@IntuitMachine

7 months

Introducing RA-DIT: Existing retrieval-augmented generation models (RAG) have shown great promise for knowledge-intensive tasks by supplementing linguistic models with relevant external information. However, current RAG methods have significant limitations. Some require…

2

65

389

Carlos E. Perez

@IntuitMachine

1 year

Convolution networks are on their last gasps of life. Facebook's Segment-Anything uses ViT models (i.e., transformers). When diffusion models migrate to ViT, it'll be completely wild!!

GitHub - facebookresearch/segment-anything: The repository provides code for running inference with...

The repository provides code for running inference with the SegmentAnything Model (SAM), links for downloading the trained model checkpoints, and example notebooks that show how to use the model. -...

github.com

4

56

384

Carlos E. Perez

@IntuitMachine

7 months

This New Research Will Change How You Design RAG Systems Should you use longer contexts or retrieval to boost your RAG model's performance on long document tasks? This important paper provides key insights through a comprehensive study comparing and combining both techniques…

8

69

378

Carlos E. Perez

@IntuitMachine

8 months

Cognitive biases, like visual illusions, are key features that characterize human-like cognition. This paper categories them into just a few. "My experience is a reasonable reference" - This belief leads to using one's own experience as an anchor for judgments about others,…

8

105

371

Carlos E. Perez

@IntuitMachine

2 months

Demis Hassabis admits that Google has some secret sauce in how Gemini is able to process 1-10m token context windows. The extreme context length in Gemini 1.5 Pro "can't" be achieved "without some new Innovations". This is an astonishing development that seems to hint at…

MachDiamonds

@andromeda74356

3 months

Many people were arguing that RAG will still be necessary in most cases even after 1 million+ long context windows because of the cost. But in a NYT interview today, Demis said they're working on caching reference materials to make subsequent processing much cheaper.

4

18

116

15

76

372