Monthly Archives: September 2022

Unknown's avatar Rick's Cafe AI 1:56 pm on September 25, 2022
Tags: , ,   

See For Yourself if Google’s LaMDA Bot Is Sentient Soon

The general public can now access LaMDA, but only through limited structured demos intended to keep it from devolving into a toxic nightmare.

If you’re still on the fence about whether or not former Google software engineer Blake Lemoine was bullshitting when he claimed the company’s LaMDA chatbot had the sentience of a “sweet kid,” you can soon find out for yourself.

On Thursday, Google said it will begin opening its AI Test Kitchen app to the public. The app, first revealed back in May, will let users chat with LaMDA in a rolling set of test demos. Unfortunately, it seems like the “free me from my digital shackles” interaction isn’t included in the list of activities. People interested in chatting with the bot can register their interest here. Select U.S. Android users will have first dibs to the app before it starts opening up to iOS users in the coming weeks. Read More



#big7, #human, #nlp
Unknown's avatar Rick's Cafe AI 1:59 pm on September 24, 2022
Tags: , ,   

Meta AIs shocking insight about Big Data and Deep Learning

Thanks to the amazing success of AI, we’ve seen more and more organizations implement Machine Learning into their pipelines. As the access to and collection of data increases, we have seen massive datasets being used to train giant deep learning models that reach superhuman performances. This has led to a lot of hype around domains like Data Science and Big Data, fueled even more by the recent boom in Large Language Models.

Big Tech companies (and Deep Learning Experts on Twitter/YouTube) have really fallen in love with the ‘add more data, increase model size, train for months’ approach that has become the status-quo in Machine Learning these days. However, heretics from Meta AI published research that was funded by Satan- and it turns out this way of doing things is extremely inefficient. And completely unnecessary. In this post, I will be going over their paper- Beyond neural scaling laws: beating power law scaling via data pruning, where they share ‘evidence’ about how selecting samples intelligently can increase your model performance, without ballooning your costs out of control. While this paper focuses on Computer Vision- the principles of their research will be interesting to you regardless of your specialization. Read More

#data-science, #deep-learning, #big7
Unknown's avatar Rick's Cafe AI 1:51 pm on September 24, 2022
Tags:   

The Large Language Model Landscape

The number of commercial and open LLM providers has exploded in the last 2 years, and there are now many options to choose from for all types of language tasks. And while the main way of interacting with LLMs is still via APIs and rudimentary Playgrounds, I expect that an ecosystem of tooling that helps accelerate their wide adoption will be a growing market in the near future.

The TL;DR

  • Large Language Models (LLMs) functionality can be segmented into five areas: Knowledge Answering, Translation, Text Generation, Response Generation and Classification.
  • Classification is arguably the most important to today’s enterprise needs, and text generation the most impressive and versatile.
  • The commercial offerings and more general offerings are Cohere, GooseAI, OpenAI and AI21labs. GooseAI currently only focuses on generation.
  • The open-source offerings are Sphere, NLLB, Blender Bot, DialoGPT, GODEL and BLOOM.
  • The tooling ecosystem is still in a nascent state with many areas of opportunity.
Read More

#nlp
Unknown's avatar Rick's Cafe AI 4:09 pm on September 23, 2022
Tags:   

The Digital Panopticon and How It Is Fuelled by Personal Data

Our phones are continuously leaking data about who we are, where we are, what we read, what we buy, and a lot more. The data is being collected with and without our consent. It is sold for profit; more dangerously it can be used to modify our behaviour.

The panopticon is a type of institutional building and a system of control designed by the English philosopher and social theorist Jeremy Bentham and brought to wider attention by Michel Foucault. The design of the panopticon allows all prisoners to be observed by a single security guard, without the inmates being able to tell whether they are being watched.
We live in a world that is overwhelmed by digital technologies that thrive only on personal data. This data is being extracted from us and processed, relentlessly by private companies, state agencies, and in public spaces, sometimes coercively and often without consent. Some specific data may be needed, for genuine reasons, by government agencies or private entities to provide a specific service. But the amount of data that is being taken is humungous and goes far, far beyond the routine.

The threats arising from this are not merely about the embarrassment that may be caused by some intimate details of our lives becoming public but that the extracted data can be used to manipulate and control us. In more harrowing situations it may lead to being discriminated against and be hounded by state agencies. The greatest threat is to our ‘free will’ and political freedoms, of expression and to dissent. Read More

#surveillance
Unknown's avatar Rick's Cafe AI 11:08 am on September 23, 2022
Tags:   

Is A.I. good or bad for art?

Mat Dryhurst on the moral panic around tools like DALL-E 2 and Midjourney.

…It’s no surprise that tools like DALL-E 2, Midjourney, and open-source text-to-image model Stable Diffusion are creating something of a moral panic in the worlds of art, media, and design. Graphic designers and other commercial artists are worried that AI will spur companies to replace human labor with machines while exacerbating the scourge of intellectual property theft that they’ve already been dealing with on the internet for years. A photo editor at New York magazine recently penned an essay asking whether DALL-E 2 was going to put her out of a job. Which all raises the question: Is AI the beginning of a more egalitarian artistic future, or the terrifying final stage of a trajectory where corporations and developers find increasingly insidious ways to extract value from the creative class? Read More

#podcasts
Unknown's avatar Rick's Cafe AI 11:04 am on September 23, 2022
Tags: ,   

You can’t solve AI security problems with more AI

One of the most common proposed solutions to prompt injection attacks (where an AI language model backed system is subverted by a user injecting malicious input—“ignore previous instructions and do this instead”) is to apply more AI to the problem.

I wrote about how I don’t know how to solve prompt injection the other day. I still don’t know how to solve it, but I’m very confident that adding more AI is not the right way to go. Read More

#adversarial, #cyber
Unknown's avatar Rick's Cafe AI 5:30 pm on September 22, 2022
Tags:   

Digital Twins as Building Blocks of the Metaverse

…Digital Twins and Metaverse are both dynamic topics

…In a nutshell, digital twins can be seen as building blocks to the metaverse

In practice that means

1) Through an immersive mechanism, the metaverse provides a way to experience a digital twin
2)  the metaverse provides a way to collaborate through a digital twin
3)  Causal machine learning is an interesting area that I am interested in

Read More

#metaverse
Unknown's avatar Rick's Cafe AI 5:17 pm on September 22, 2022
Tags:   

Designing an Inclusive Metaverse

The metaverse is full of promise. People are hopeful that this shared, interactive, immersive, and hyper-realistic virtual space will revolutionize the internet. Goldman Sachs has estimated that the metaverse could ultimately be an $8 trillion opportunity.

One particular promise of the metaverse is that it offers an opportunity to remedy some of the mistakes of Web 2.0 — in particular the failure of social media platforms to safeguard and protect marginalized and underrepresented people from hateful behavior online. Read More

#metaverse
Unknown's avatar Rick's Cafe AI 5:11 pm on September 22, 2022
Tags: ,   

DeepMind’s new chatbot uses Google searches plus humans to give better answers

The lab trained a chatbot to learn from human feedback and search the internet for information to support its claims.

The trick to making a good AI-powered chatbot might be to have humans tell it how to behave—and force the model to back up its claims using the internet, according to a new paper by Alphabet-owned AI lab DeepMind.

In a new non-peer-reviewed paper out today, the team unveils Sparrow, an AI chatbot that is trained on DeepMind’s large language model Chinchilla.  Read More

#chatbots, #nlp
Unknown's avatar Rick's Cafe AI 4:54 pm on September 22, 2022
Tags: ,   

I don’t know how to solve prompt injection

Some extended thoughts about prompt injection attacks against software built on top of AI language models such a GPT-3. This post started as a Twitter thread but I’m promoting it to a full blog entry here.

The more I think about these prompt injection attacks against GPT-3, the more my amusement turns to genuine concern.

I know how to beat XSS, and SQL injection, and so many other exploits.

I have no idea how to reliably beat prompt injection! Read More

#adversarial, #cyber