AI — Страница 1550

ServiceNow Introduces ‘Innovation Centre’ in India, Looks to Transform Work with Generative AI

Digital workflow company ServiceNow on Monday announced the opening of its digital incubation hub – “Innovation Centre” for Indian enterprises to model, demonstrate and stress-test digital roadmaps, such as GenAI, hyper automation and low code apps to deliver a better experience for customers and employees.

Located in Hyderabad, Knowledge City, the Innovation Centre boasts nine purpose-built rooms dedicated to assisting companies in creating customized digital blueprints to scale their business operations. In addition to its primary function, the centre will facilitate on-demand training sessions in collaboration with academia, customers, and partners committed to enhancing the digital skills of Indian citizens.

Establishing an Innovation Centre in Hyderabad, enables our customers to have conversations directly with our engineers at ServiceNow’s largest development centre outside of the USA, and integrate emerging technologies customised to their business strategies to best meet their needs.” said Kamolika Gupta Peres, Vice President & Managing Director, ServiceNow Indian Sub-Continent.

Furthermore, ServiceNow’s global RiseUp program has partnered with ten academic and government institutions in India over the past ten months, with a commitment to train over 5,500 students in digital skills during the initial phase.

Recently, ServiceNow introduced ‘Now Assist for Virtual Agent’, a cutting-edge solution that enhances the capabilities of the platform’s chatbot through the application of Generative AI in textual scenarios.

The post ServiceNow Introduces ‘Innovation Centre’ in India, Looks to Transform Work with Generative AI appeared first on Analytics India Magazine.

ChatGPT-Powered Data Exploration: Unlock Hidden Insights in Your Dataset

Image by Author

Analyzing a dataset to understand it better and gain insights is an important step in any data science project. And ChatGPT can help simplify and speed up such analysis.

Here we’ll perform exploratory data analysis on a sample dataset using ChatGPT. We’ll understand distributions of features, how a feature of interest varies with other features, and much more.

To follow along and access ChatGPT, you need an OpenAI account. If you don't have one, you can sign up for a free account.

Loading the Dataset and Getting Basic Info

The seaborn library has some ready-to-use datasets that we can analyze. We’ll use the tips dataset.

Prompt: Tell me more about the `tips` dataset built into the seaborn library.

Knowing what the dataset is about is a good starting point. Let's start by doing the following:

Load the dataset into a pandas dataframe, and
Get basic info and summary statistics of the columns to understand more about the dataset.

Prompt: Let’s use the `tips` dataset. Write the code to read the dataset into a pandas dataframe and get the basic info and summary statistics of the available fields.

import seaborn as sns  import pandas as pd    # Load the 'tips' dataset from Seaborn  tips_data = sns.load_dataset('tips')    # Create a Pandas DataFrame from the loaded dataset  tips_df = pd.DataFrame(tips_data)    # Display the first few rows of the DataFrame  print("First few rows of the 'tips' dataset:")  print(tips_df.head())    # Get basic information about the fields  print("nInformation about the 'tips' dataset:")  print(tips_df.info())    # Get summary statistics of the numeric fields  print("nSummary statistics of the numeric fields:")  print(tips_df.describe())

We see that we have quite an informative summary of the dataset:

First few rows of the 'tips' dataset:     total_bill   tip     sex smoker  day    time  size  0       16.99  1.01  Female     No  Sun  Dinner     2  1       10.34  1.66    Male     No  Sun  Dinner     3  2       21.01  3.50    Male     No  Sun  Dinner     3  3       23.68  3.31    Male     No  Sun  Dinner     2  4       24.59  3.61  Female     No  Sun  Dinner     4    Information about the 'tips' dataset:    RangeIndex: 244 entries, 0 to 243  Data columns (total 7 columns):   #   Column      Non-Null Count  Dtype     ---  ------      --------------  -----      0   total_bill  244 non-null    float64    1   tip         244 non-null    float64    2   sex         244 non-null    category   3   smoker      244 non-null    category   4   day         244 non-null    category   5   time        244 non-null    category   6   size        244 non-null    int64     dtypes: category(4), float64(2), int64(1)  memory usage: 7.4 KB  None    Summary statistics of the numeric fields:         total_bill         tip        size  count  244.000000  244.000000  244.000000  mean    19.785943    2.998279    2.569672  std      8.902412    1.383638    0.951100  min      3.070000    1.000000    1.000000  25%     13.347500    2.000000    2.000000  50%     17.795000    2.900000    2.000000  75%     24.127500    3.562500    3.000000  max     50.810000   10.000000    6.000000

From the summary statistics, we have an idea of the numerical features in the dataset. We know the minimum and maximum values, mean and median values, and percentile values for the numerical features. There are no missing values so we can proceed with the next steps.

Exploring the Dataset – The What, the Why, and the How

Now that we have an idea of the dataset, let's go further.

The goal of this exploratory data analysis is to understand the tipping behavior better. For this we can come up with helpful visualizations. These should help us understand the relationship of the tip amount to the various categorical variables in the dataset.

Because this is a simple dataset to analyze, let's prompt ChatGPT to give us a set of steps to go about analyzing this data set further.

Prompt: The goal of the analysis is to get some insights into the tipping behavior, starting with some simple visualizations. Suggest steps for the same.

The data exploration steps suggested by ChatGPT all seem valid. So we will use these steps—to better understand a dataset—one step at a time. We’ll prompt ChatGPT to generate code, try to run the generated code, and modify them as needed.

Exploring the Distribution of Tip Amounts

As a first step, let's visualize the distribution of the tip amount prompt.

Prompt: Write the code to plot the distribution of tip amounts.

ChatGPT gave the following code to generate a histogram and a kernel density plot that help us get an idea of the distribution of the tip amounts:

import matplotlib.pyplot as plt    # Create a histogram of tip amounts  plt.figure(figsize=(8, 6))  sns.histplot(data=tips_data, x='tip', kde=True)  plt.title("Distribution of Tip Amounts")  plt.xlabel("Tip Amount")  plt.ylabel("Frequency")  plt.show()

Analyzing Tipping Behavior Based on Categorical Variables

We have many categorical variables in the dataset. It will help to have a simple bar plot showing the average tip amount for the different values of these category variables.

Prompt: Use all the available categorical variables (‘sex’, ‘smoker’, ‘day’, and ‘time’). Write the code to generate bar plots to understand tipping behavior. Use subplots so we can have all the four bar plots in a single figure.

And here’s the code. The code ran without errors; I only had to modify the figure size:

# Define the categorical variables to analyze  categorical_vars = ['sex', 'smoker', 'day', 'time']    # Create subplots for each categorical variable  fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(8, 5))  fig.suptitle("Tipping Behavior based on Categorical Variables", fontsize=16)    # Generate bar plots for each categorical variable  for ax, var in zip(axes.flatten(), categorical_vars):      sns.barplot(data=tips_data, x=var, y='tip', ax=ax)      ax.set_xlabel(var.capitalize())      ax.set_ylabel("Average Tip Amount")    plt.tight_layout()  plt.show()

From the plots, we see that features like sex and smoking behavior don’t influence tipping behavior (which is expected). While days and times seem to. The average tip amount on weekends and dinner seem to be slightly higher.

Visualizing the Relationship Between Total Bill and Tip Amount

Now, let’s see how the total bill influences the tip amount paid.

Prompt: I’d like to understand the relationship between total bill and the tip amount. Please give me the code to generate a suitable plot for this. I believe a simple scatter plot will be helpful.

Here’s the code to generate the required scatter plot:

# Create a scatter plot of total bill vs. tip amount  plt.figure(figsize=(6, 4))  sns.scatterplot(data=tips_data, x='total_bill', y='tip')  plt.title("Total Bill vs. Tip Amount")  plt.xlabel("Total Bill")  plt.ylabel("Tip Amount")  plt.show()

We see that there’s some positive correlation between the total bill and the tip amounts.

Visualizing the Relationship Between Party Size and Tip Amount

As a next step, let’s try visualizing the relationship between the party size and the tip amount.

Prompt: Now, I'd like to understand how the tip amount varies with the number of people who dine (party size). Which is a better choice: a bar plot or a violin plot?

As we’d also like to understand the distribution of the tip amounts, let’s go ahead with creating a violin plot.

Prompt: Cool! Please write the code to generate a violin plot for this visualization.

And here’s the code:

# Create a violin plot for tip amount by party size  plt.figure(figsize=(6, 4))  sns.violinplot(data=tips_data, x='size', y='tip')  plt.title("Tip Amount by Party Size")  plt.xlabel("Party Size")  plt.ylabel("Tip Amount")  plt.show()

Understanding Tipping Behavior Based on Both Time and Day

Next, let’s understand how the time and the day influence the tipping behavior.

Prompt: I want to understand how the tipping behavior varies depending on both time and day. What is a good visualization? Also, write the code to generate the required visualization.

This code snippet shows how to pivot using the day as the index and aggregate the mean value of the tip amounts for lunch and dinner (the time of the day):

# Create a pivot table of average tip amount by time and day  pivot_table = tips_data.pivot_table(values='tip', index='day', columns='time', aggfunc='mean')    # Create a heatmap of tipping behavior based on time and day  plt.figure(figsize=(8, 6))  sns.heatmap(pivot_table, cmap='YlGnBu', annot=True, fmt=".2f", cbar=True)  plt.title("Tipping Behavior based on Time and Day")  plt.xlabel("Time")  plt.ylabel("Day")  plt.show()

From the heat map, we see that the average tip amounts are marginally higher for dinner on Sunday than other time-day pairs. In general, for a given day, on an average, tip amounts for dinner seem to be higher than those for lunch.

We also see that there is not enough data for lunch on Saturdays and Sundays. Because we only have a total of 244 records, we may not have any record for lunches on weekends. Let’s go ahead and verify this.

Prompt: I want to get the number of parties who ordered lunch on weekends. Write the code to filter the `tips_data` dataframe and get this count.

Here’s the code to filter the dataframe and obtain a subset containing the records for lunches on weekends:

# Filter the data for lunch on weekends  weekend_lunch_data = tips_data[(tips_data['time'] == 'Lunch') & (tips_data['day'].isin(['Sat', 'Sun']))]    # Get the count of parties who ordered lunch on weekends  count = weekend_lunch_data['size'].count()  print("Number of parties who ordered lunch on weekends:", count)

And we see that we don’t have any records for the same, so our analysis is correct thus far:

Number of parties who ordered lunch on weekends: 0

And that’s a wrap! We explored the `tips` dataset and generated some helpful visualizations by prompting ChatGPT.

Wrapping Up

In this article, we learned how to leverage ChatGPT for data exploration. If you’re interested in integrating ChatGPT into your data science workflow, check out this guide. It walks through an example project—along with tips and best practices—to effectively use ChatGPT for data science experiments.
Bala Priya C is a developer and technical writer from India. She likes working at the intersection of math, programming, data science, and content creation. Her areas of interest and expertise include DevOps, data science, and natural language processing. She enjoys reading, writing, coding, and coffee! Currently, she's working on learning and sharing her knowledge with the developer community by authoring tutorials, how-to guides, opinion pieces, and more.

Top 8 Use Cases of Bard’s New Image Recognition

Ever since Google integrated the powers of Google Lens into Bard, enabling users to inquire the chatbot for insights about an image or create descriptive captions, the image recognition game went through a massive makeover. Not only can it identify the things in the picture, but it can also extract text and make sense of the image. Even GPT-4 is multimodal. But it is still in the research phase. Several users have taken to Twitter to express their excitement about the new update.

Let’s take a look at some interesting use cases of the same.

Read and Understand Images

Let’s start with the basics. Bard is now able to understand and explain a picture. I uploaded a picture of Salvador Dali’s masterpiece ‘The Persistence of Memory’ and asked Bard to explain its meaning. The chatbot quickly gave detailed descriptions of the paintings, and the story behind them and also told me that it can be interpreted in different ways.

Another user shared that she gave an image of a pug wearing a graduation cap and asked Bard about what is happening. It gave three potential instances of how the puppy could be graduating from obedience school or therapy program and more.

Create a Website from Sketch

Add an image of your sketch with the ‘+’ option and provide your prompt, for example, “Compose a concise HTML/JS script to transform this mock-up into a vibrant website, wherein the jokes are substituted with two genuine jokes.” Since the initial outcome may not meet user expectations, Bard offers additional choices by selecting “View other drafts.” Alternatively, you can regenerate the output. If any specific modifications are desired, you can make another request in a separate prompt. To execute the script, the HTML code should be copied into a text editor or saved as a text file.

Understand Complex Graphs

Another use used Google Bard’s image input function to read text and graphs on the GPT-4 demo slides and perform calculations and that too in Japanese. They explain that they used to rely on ChatGPT or Perplexity for their tasks, but Bard has now become a powerful tool for them, depending on the specific application.

Google Bardの画像入力機能で、GPT-4のデモであったスライドの文字とグラフを読み取って計算する、というのも余裕でできてる…！
少し前までBardのポジションが中途半端でChatGPTかPerplexityしか使ってなかったんだけど、用途によってはBardがかなり強力なツールに躍り出た…！ pic.twitter.com/QHUG9A86DK

— KAJI | 梶谷健人 (@kajikent) July 14, 2023

Cooking Gets Easier

Saw a picture of pasta on your feed and now you have a craving for it? Well, now you can upload an image of a meal and ask for a full recipe, Bard will give it to you. And that is what AI influencer Rowan Cheung did.

Create iPhone App from a Screenshot

Ammar Barshi, the design manager at Brex, used Bard to replicate a basic timer application for the iPhone, in just 4 minutes from a screenshot without any explicit cues about the app’s functionality. Bard generated the necessary code, although it did commit a few errors, but they were easily rectified.

Brain CT Diagnoses

Another user added the image of a CT scan and asked Bard to understand it. The user gave inputs in Japanese. Bard was able to list the potential causes despite lacking specialised expertise in the field.

Googleのbard凄い & 怖い！！！
　
Bardによる脳CTの診断！
　
｢画像には脳の白い部分が写っていると言えます。。。脳腫瘍、脳卒中、出血など、さまざまな原因が考えられます｣
　
答えは脳出血です。その鑑別診断を挙げるだけでも凄いですね。特化型で学習しているわけではありませんので。 pic.twitter.com/aEdF5xtlqt

— 河野健一　生成AI ✕ 医療に注目！手術支援AI CEO 脳外科医 (@CeoImed) July 14, 2023

Social Media Caption

Bard is also a great buddy if you want to have great captions for your social media posts. It can generate captions according to your needs.

Personal Expense Manager

If you need to compile your expenditure for an expense report but you have too many bills to handle, you can now use Bard’s image recognition to take photos of the receipts and feed them to the chatbot. Bard organises the receipts into a table with details like date, time, category, description, and amount and this table can be exported to Google Sheets, eliminating the need for separate expense report apps.

The post Top 8 Use Cases of Bard’s New Image Recognition appeared first on Analytics India Magazine.

Apple’s Vision Pro Woes

Apple Vision Pro, the much-awaited VR headset touted as the company’s most important product release since the iPhone, was recently unveiled after seven years of development. Apple announced that the “spatial computing” headset, priced at $3,500, will not be available for sale until early next year, a considerable delay from its June 5 launch where 1 million units were promised before 2024. Analysts believe this delay is primarily due to supply chain issues rather than providing developers with ample time to create apps for the Vision Pro.

Manufacturing Malfunctions

Multiple industry sources have confirmed that Luxshare is currently the sole assembler for Apple’s device. Additionally, two Chinese suppliers of specific components for the Vision Pro have stated that Apple has only requested enough supplies for 130,000 to 150,000 units in the first year.

These projections indicate a significant reduction in production compared to the initial target of 1 million units in the first 12 months. Analysts and industry experts interpret these low volume forecasts as a lack of confidence from Apple in scaling up production, following a history of missed deadlines in launching the device.

One of the significant challenges faced in production is the manufacturing of the device’s sleek screens. These screens consist of two micro-OLED displays, one per eye, and an outward-facing curved “lenticular” lens. The internal displays offer a resolution surpassing anything currently available in the market, while the outward lens projects the wearer’s eyes to the surrounding environment.

Jay Goldberg, founder of tech consultancy D/D Advisors, acknowledged that these challenges are part of the normal growing pains experienced in developing the most complex consumer device to date. He said, “There is a lot of technology in the Vision Pro and they knew it would take a while to scale up. The company knows they won’t make money on this in the first year.”

He also noted that the higher-than-expected price of $3,500 indicates that Apple accounted for production inefficiencies, knowing that manufacturing yields would be lower compared to its established product lineup.

The reduction in 2024 production forecasts has disappointed Luxshare, which had prepared its capacity to assemble nearly 18 million units annually in the coming years, according to a source familiar with the company. Analysts have observed that the Vision Pro has not provided a substantial boost to the headset supply chain in Asia. Eddie Han, an analyst at Taiwan-based Isaiah Research, stated that Apple has not delivered a superior product as anticipated, leading to low confidence among manufacturers.

Wall Street analysts’ predictions for Vision Pro sales vary widely, ranging from several hundred thousand to several million units in the first year. At the time of the headset’s unveiling, Wedbush estimated that Apple would ship around 150,000 units in the first year, while Morgan Stanley projected approximately 850,000 units, and Goldman Sachs believed shipments could reach 5 million in 2024. For comparison, Apple sold 1.4 million iPhones in its first year on the market.

In the past as well, Apple has seen delays in manufacturing, with their Micro-LED displays in the Apple watch which caused manufacturing bottlenecks. This time they’re facing significant challenges with its mixed-reality Vision Pro headset due to the complexity of the headset’s design and production. Apple has had to revise their targets and delay the release of the product in Canada, UK and Asia.

The post Apple’s Vision Pro Woes appeared first on Analytics India Magazine.

Automating the Chain of Thought: How AI Can Prompt Itself to Reason

Image created by author with Midjourney Key Points

Chain-of-thought (CoT) prompting improves LM reasoning by providing step-by-step examples
Manual creation of CoT demonstrations requires non-trivial human effort
This paper explores automating CoT demonstration generation using the LM itself
The proposed Auto-CoT method clusters questions then samples diverse ones for self-prompting
Experiments show Auto-CoT matches manually created CoT, without human involvement

Introduction

The paper "Automatic Chain of Thought Prompting in Large Language Models" explores automated ways to create effective "chain of thought" (CoT) prompts for large language models (LLMs) like GPT-4. CoT prompting involves showing the LLM examples that demonstrate step-by-step reasoning chains mapping from a question to a final answer. This improves performance on complex reasoning tasks.

Discussion

The best CoT prompting results, however, currently require humans to manually create demonstrations, with hand-crafted questions and detailed reasoning steps tailored to each task. The authors propose eliminating this manual effort by having the LLM automatically generate its own CoT demonstrations for prompting. Their key method, called Auto-CoT, works by first clustering the questions of a given task based on their semantic similarity. Auto-CoT then samples a diverse set of questions covering different clusters. For each sampled question, Auto-CoT uses the LLM itself in zero-shot mode to produce a reasoning chain from the question to an answer. It applies simple heuristics to select chains based on length and simplicity.

The authors perform experiments evaluating Auto-CoT on 10 reasoning datasets spanning arithmetic, common sense, and symbolic logic problems. The results show that Auto-CoT matches or exceeds the performance of CoT prompting based on manually created demonstrations, without requiring any human effort to design demonstrations. A key insight is that using diversity-based sampling over similarity-based retrieval to select the prompting questions mitigates the impact of imperfect demonstrations generated by the LLM's zero-shot reasoning. Auto-CoT also substantially outperforms baselines like retrieving similar questions or random sampling for the demonstrations.

Overall, the work provides strong evidence that LLMs can prompt themselves to demonstrate complex multi-step reasoning. Auto-CoT essentially composes one LLM that generates a diverse set of CoT examples, with another LLM that uses those examples for inference. The authors suggest this self-prompting approach could significantly extend prompting techniques and make LLMs much better few-shot learners on complex reasoning tasks. Limitations include potential computational costs and issues scaling to more unconstrained problems. But the ability to automate prompting reduces human effort and customization needs.

Research Q&A

How does Auto-CoT compare to other methods that automate prompt creation, like retrieval-augmented prompting?

Retrieval-augmented prompting retrieves related data examples to use for prompting, rather than having the LLM generate demonstrations. A key difference is that Auto-CoT doesn't require a dataset of labeled examples and instead relies on the LLM's own zero-shot reasoning. Retrieval may be more sample-efficient but requires data collection. Auto-CoT is fully automated but can suffer from imperfect demonstrations.

Could Auto-CoT be applied to natural language generation tasks beyond logical reasoning?

The clustering and self-prompting approach seems promising for less structured textual tasks where coherence is important. For example, Auto-CoT could provide writing planning examples for creative writing, or dialog illustrations for conversational bots. The key challenges would be defining appropriate clustering methods and training the LLM's zero-shot generation for high-quality demonstrations.

What is innovative about this research?

The key innovation is using the LLM itself to generate demonstrations for prompting, instead of relying on manual creation. This allows prompting to become more automated and task-adaptive. The clustering to select diverse questions for self-prompting is also innovative.

What are the broader implications of this research?

This research could significantly reduce the human effort and expertise needed to design effective prompts. It may allow LLMs to learn new tasks more quickly and from less data, enhancing their few-shot learning capabilities. The self-prompting approach could be applied to extend prompting techniques like in-context learning.

What are some potential issues or oversights with this research as presented, if any?

A potential issue is that Auto-CoT relies on clustering questions based on similarity features from Sentence-BERT. Performance could suffer on tasks where semantic similarity doesn't align well with reasoning similarity. The approach also likely incurs higher compute costs than standard prompting.

What are the logical next research steps from this research?

Important next steps include exploring how Auto-CoT scales to more complex and open-ended reasoning tasks, integrating it with retrieval of external knowledge sources, and studying if the approach can be learned more sample-efficiently through meta-learning rather than relying solely on a pre-trained LLM. Analyzing the interplay between cluster count, sample size, and performance is also an open question.

Takeaways

Auto-CoT reduces the need for hand-crafted demonstrations to prompt LMs
Self-prompting with Auto-CoT composes one LM generating diverse examples, and another inferring
Diversity in sampling questions is key to overcoming imperfect zero-shot reasoning chains
The approach could extend prompting techniques and make LMs better few-shot learners
Auto-CoT demonstrates the promise of automating prompting to reduce human effort
Next steps include scaling Auto-CoT to more complex reasoning tasks and larger LMs

Matthew Mayo (@mattmayo13) is a Data Scientist and the Editor-in-Chief of KDnuggets, the seminal online Data Science and Machine Learning resource. His interests lie in natural language processing, algorithm design and optimization, unsupervised learning, neural networks, and automated approaches to machine learning. Matthew holds a Master's degree in computer science and a graduate diploma in data mining. He can be reached at editor1 at kdnuggets[dot]com.

Indian-Origin Researcher’s AI Breakthrough Simplifies Video Translation Across Languages

Indian origin researcher Prady Modukuru has made a remarkable tool that effortlessly translates any video into the language of your choice. All you need is a YouTube link!!

Modukuru posted a video on Twitter translating a conversation between Meta CEO Zuckerberg and Lex Fridman to Hindi. Here is the glimpse.

I built the simplest way to convert your video to other languages.
all you need is a youtube link. no install required.
curious? check out @lexfridman + zuck talking about @elonmusk in hindi. pic.twitter.com/nZBi4ZtRde

— Prady (@therealprady) July 16, 2023

A serial entrepreneur, Modukuru is currently working on a stealth startup. Some of his previous ventures include marketing startup sunya and mental wellness startup Oneself (inactive). Previously, he has worked with companies like Microsoft, Cisco, Qknow among others.

How it works?

Modukuru leveraged the power of GPT-4 and prompt engineering techniques for seamless translation. To ensure accurate voice representation in different languages while preserving the original speaker’s voice, he collaborated with ElevenLabs for voice training and text-to-speech capabilities

Moreover, for achieving precise lip syncing, Modukuru has also utilised wav2lip-2, which harnesses AI for generating lifelike lip movements. This breakthrough in the AI industry opens up new possibilities, allowing users to easily comprehend informative content in their preferred language.

In the near future, this technology can revolutionise dubbing, and can be used by major OTT platforms to translate content including series and movies to any other language.

The post Indian-Origin Researcher’s AI Breakthrough Simplifies Video Translation Across Languages appeared first on Analytics India Magazine.

Meta Needs You for Their Generative AI

Meta wants you to help them with their generative AI initiatives and is all out conducting community programs to achieve it.

In November last year, Meta announced that it would run Community Forums as a way to help the company make decisions on their technologies. By allowing a diverse group of people to discuss issues, and offer perspectives and recommendations, Meta believes that this would ‘improve the quality of governance.’ Meta’s focus at that time was Metaverse.

Collaborating with Stanford University, the results of the first global deliberative poll was released last month, which involved 6300 people from 32 countries and nine regions around the world. The participants spent hours in conversations via online group sessions and interacted with non-Meta experts about the issues under discussion. The topic : moderation and monitoring systems for bullying and harassment in the metaverse. Months of experiment and ironically, metaverse is no longer relevant. However, 82% of the participants recommended the same deliberative democracy format be followed by Meta for making future decisions — Meta has decided to follow a similar process for their generative AI tech.

Humans in the Loop

Quite literally, keeping people in the loop for decision-making is Meta’s new model. Last month, the company launched a Community Forum on Generative AI with the goal of gaining feedback on what people would ‘want to see reflected in new AI technologies’. Meta believes in incorporating people and experts’ say in product and policy decisions around generative AI, and they claim to be actively working with academics, researchers and community leaders. But, why the push?

Having faced enough flak in the past around capturing user information and breaching data privacy on social media platforms (Facebook and Instagram), Mark Zuckerberg must be probably pulling a reverse move by seeming to give control to people for formulating the next step.

Meta is also a founding member of Partnership on AI, a non-profit community, since 2016, where they work with industry experts, organisations, media and others to address concerns about the future of AI and to formulate ‘right ethical boundaries.’ Ironically, Meta’s recently launched microblogging platform Threads, coerces users to give access to personal information on the phone in order to use the app.

Not The Best Approach

The human-feedback system that Meta is experimenting does come with its limitations. How much of people’s feedback is flawless and how much of it can be implemented in the system is something that is questionable. On their pilot community program for mitigating bullying in metaverse, the participants were not aligned with punishing the users involved in repeated bullying and harassment. For instance, removing members-only spaces that saw repeated bullying had only 43% support.

Furthermore, the participants had no interaction with the decision makers i.e. Meta employees which made the process seem like a simple survey or an experiment on data-gathering rather than a democratic exercise.

In Others I Trust

With the countless talks that’s surfacing around AI safety guidelines and the need for universal regulatory policies, every major tech is claiming to work towards it. Meta is no exception in following another tech company’s work – OpenAI.

Meta is trying hard to catch up with OpenAI and speeding ahead in the open source LLM race. Tracing the reigning chatbot maker’s path, Meta seems to be going through the democratic decision-making policy that OpenAI is pursuing.

OpenAI had announced grants with $1 million to fund experiments to democratise AI rules and best solve AI safety mishaps. The company also announced another million for their cybersecurity grant program, for creation and advancement of AI-powered cybersecurity tools and technologies. In other words, a program where people can help create/fix the company’s security framework.

While the move can be critically looked at as a tactic to avoid the government from interfering with the company’s plan for AI regulation, or even as a way to seem like a responsible company working ‘for the people,’ big tech is slowly adopting the democratic route.

Recently, Anthropic spoke about how they would work better on constitutional AI by talking to people and not just experts. DeepMind, recently released a paper that addresses and investigates how international institutions can help manage and mitigate AI risks. In one of the complementary models that the company proposed, an AI safety project aims to bring together researchers and engineers to access advanced AI models for research.

The post Meta Needs You for Their Generative AI appeared first on Analytics India Magazine.

Meta AI Releases A Multimodal Model “CM3leon” — But Won’t Release It

Meta AI, the AI arm of Mark Zuckerberg owned Meta, has introduced a multimodal model “CM3leon” (pronounced like chameleon), that does both text-to-image and image-to-text generation. Similar to its previous models, the company has decided not to release the model code which has enraged AI enthusiasts.

The AI model bridges the gap between text and images. With its capabilities in text-guided image generation and editing. The model has the potential to revolutionise the approach in which current users interact with and manipulate visual content.

According to Meta, its model boasts improved capabilities in producing coherent imagery that closely aligns with input prompts. What sets CM3leon apart is its efficiency, as it requires only five times the computing power and a smaller training dataset compared to previous transformer-based methods.

Meta touts CM3leon’s prowess in various vision-language tasks, including visual question answering and long-form captioning. This novel approach by Meta marks a departure from the diffusion method commonly employed in image generation. Instead, the company’s researchers opted for the Transformer architecture, a neural network design widely recognized for its successful implementation in large language models like OpenAI’s GPT-4. While CM3leon is not the first transformer-based image generator—StyleSwin precedes it—Meta asserts that it surpasses other contenders in terms of efficiency.

Though the model is being lauded for being state-of-the-art the fact that it is closed source makes the research community irk. Similarly, last month the tech giant unveiled Voicebox but did not make the model public over fears of potential misuse. “While we believe it is important to be open with the AI community and to share our research to advance the state of the art in AI, it’s also necessary to strike the right balance between openness with responsibility,” Meta had stated while releasing the paper.

Again, Meta is not the only tech goliath gatekeeping its research. Even though companies like Google, Microsoft, Facebook, Amazon, and Apple have contributed tons to research over the years — they also have kept a major chunk of it to themselves. The reason this displeases the research community is because these companies have been dependent on the open source community yet refuse to contribute as much as they gain.

Read more: Big Techs Flip-Flop on Open Source

The post Meta AI Releases A Multimodal Model “CM3leon” — But Won’t Release It appeared first on Analytics India Magazine.

8 Outstanding Papers Presented at the ACL 2023

The 61st chapter of the Association for Computational Linguistics is ongoing in Toronto, Canada bringing together researchers and practitioners from the field of computational linguistics.

Out of the plethora of research showcased at the annual meet, we have picked out the 8 brilliant papers at the conference that caught our attention!

Backpack Language Models

AI language models exhibit gender bias in pronoun distributions, favoring gendered pronouns based on context. This bias can be flipped by replacing stereotypically associated professions. However, achieving consistent debiasing across all contexts is challenging.

Backpack LM addresses this by leveraging non-contextual sense vectors, capturing multiple aspects of a word’s meaning. By incorporating Backpack LM, we can mitigate biases and create fairer, more inclusive language models with improved interpretability and control.

Authors: John Hewitt, John Thickstun, Christopher D. Manning, Percy Liang

Do Androids Laugh at Electric Sheep? Humor “Understanding” Benchmarks from The New Yorker Caption Contest

Can AI models truly grasp humor? In the research paper, resarchers test them with New Yorker Cartoon Caption Contest tasks: matching jokes to cartoons, identifying winning captions, and explaining their humor.

The authors explored the capabilities of both multimodal models, which directly engage with cartoon images, and language-only models, which are provided with rich descriptions.

Authors: Jack Hessel, Ana Marasovic, Jena D. Hwang, Lillian Lee, Jeff Da, Rowan Zellers, Robert Mankoff and Yejin Choi

Don’t Generate, Discriminate: A Proposal for Grounding Language Models to Real-World Environments

Current language models lack the ability to ground to real-world environments. Existing approaches place the burden of generating executable plans on the language models themselves, which leads to challenges in maintaining grammaticality, faithfulness, and controllability.

To address this, in this paper, researchers introduce Pangu, a framework that leverages the discriminative power of language models for grounded language understanding, instead of relying on their generative capabilities.

Authors: Yu Gu, Xiang Deng and Yu Su

Minding Language Models’ (Lack of) Theory of Mind: A Plug-and-Play Multi-Character Belief Tracker

Large-scale neural language models lack basic Theory of Mind (ToM) — the ability to reason about the mental states of other people.

Researchers propose SymbolicToM, a plug-and-play approach that enables reasoning about belief states of multiple characters using explicit symbolic representation. It tracks each entity’s beliefs, estimations of others’ beliefs, and higher-order reasoning through graphical representations, enhancing precision and interpretability in reading comprehension tasks.

Authors: Melanie Sclar, Sachin Kumar, Peter West, Alane Suhr, Yejin Choi and Yulia Tsvetkov

The Mechanical Bard: An Interpretable Machine Learning Approach to Shakespearean Sonnet Generation

Researchers explore automated generation of Shakespearean sonnets, utilizing constrained decoding to adhere to meter, rhyme scheme, length, and poetic conventions.

The approach produces sonnets resembling human-authored ones, with lyrical language, literary devices, and adherence to genre constraints, as confirmed by human evaluation.

Authors: Edwin Agnew, Michelle Qiu, Lily Zhu, Sam Wiseman and Cynthia Rudin

World-to-Words: Grounded Open Vocabulary Acquisition through Fast Mapping in Vision-Language Models

Grounding language in the physical world is crucial for understanding word meanings. To introduce the factor in language models, researchers present Grounded Open Vocabulary Acquisition (GOVA), which explores grounding and bootstrapping in open-world language learning.

Their initial approach is object-oriented BERT (OctoBERT), a visually-grounded language model that pre-trains on image-text pairs with a focus on grounding.

Authors: Ziqiao Ma, Jiayi Pan and Joyce Chai

Forgotten Knowledge: Examining the Citational Amnesia in NLP

Have you ever wondered how old are the papers you cite? Or If we fail to read older papers and benefit from important ideas? In this paper, researchers explore questions like these about Natural Language Processing (NLP) papers with data and graphs.

Authors: Janvijay Singh, Mukund Rungta, Diyi Yang and Saif Mohammad

Causes and Cures for Interference in Multilingual Translation

This research paper from Meta explores the little-understood phenomenon of interference, broadly defined as a negative interaction between different translation directions in a multilingual machine translation model.

“Interference trends can be tricky to measure,” lead author Uri Shaham acknowledged in a December 16, 2022 tweet, summing up the paper’s central questions: “What causes interference or synergy between language pairs in multilingual translation? Do we actually need specialized algorithms to alleviate interference?”

Authors: Uri Shaham, Maha Elbayad, Vedanuj Goswami, Omer Levy, Shruti Bhosale

The post 8 Outstanding Papers Presented at the ACL 2023 appeared first on Analytics India Magazine.

Flyfish Wants to be a Premium Indian ‘Consultative Sales AI’ Offering

Fractal, a global provider of AI and advanced analytics solutions, has introduced Flyfish, a 360-degree generative AI platform for digital sales. The platform aims to enhance revenue growth and foster long-lasting customer relationships.

The vision behind Flyfish is to revolutionise the sales experience by making technology more human-like. Flyfish offers consultative experiences through intuitive sales advisors, enabling personalised, data-driven shopping interactions with consumers.

The platform aims to create a new market category called consultative sales AI. By offering personalised guidance and assistance to customers, Flyfish enhances the customer experience and boosts sales for businesses. According to the team, the platform has shown promising results in pilot programs, with increases in lead generation, conversion rates, and a significant reduction in customer acquisition costs.

“In the close-door pilots that we have built and even launched for some of our customers, we measured it and there is a significant jump, in lead generation there was a 35% jump in in one of the cases, and then conversion rate went up by 12% in another case,” Shridhar Marri CEO and co-founder of Flyfish said during a recent interview with AIM.

He also predicted that the product would help enterprises bring customer acquisition costs down by 40% over a period of three years.

How it Works

Flyfish has a preprocessor that analyses user questions or context, determines the appropriate flow, and retrieves relevant products and recommendations. This information is sent to a language model (LLM) to generate a humanised response, which is then post-processed within Flyfish. The platform also includes a channel connector that allows integration with various digital channels like web, messaging platforms, and social media.

If needed, a human can be brought into the conversation, and their responses are delivered through the same channels. Custom flows can be created within Flyfish for specific enterprises, such as incorporating sponsored or complementary product options.

Flyfish can also be used for marketing, where it can provide a point-of-sale or advisory experience for users clicking on ads. The platform is agnostic to the choice of language models, allowing users to select their preferred LM, such as GPT, Google’s models, or others. Flyfish integrates with messaging APIs to connect with different platforms like WhatsApp, Telegram, and more.

The platform seamlessly integrates into existing sales and marketing tech stacks, encompassing e-commerce platforms, product catalogues, pricing, availability, and more. It operates on a cloud-agnostic architecture, utilising various cloud services such as Azure, ensuring flexibility and scalability.

Addresses Specific Consumer Needs

Flyfish addresses the diverse needs and contexts of individual consumers, streamlining their research process and significantly reducing the time required to make informed purchasing decisions. By leveraging generative AI, Flyfish delivers a range of benefits, including the ability to provide tailored product recommendations and suggestions that consumers may not have discovered otherwise.

“I think we’re going to change the way consumers interact with the brands. Every consumer, if you look at them, they have different needs and contexts.”

He explained that, when making a significant purchase like a car, it is common for consumers to go through extensive research that can last anywhere from six to eight months. This research period applies to various types of purchases, with varying time frames depending on the specific item. Regardless of the product, consumers invest significant time in researching their options, often spanning across days, weeks, or even months.

Marri expects Flyfish to reduce this time and result in an enjoyable experience.

A Secure Enterprise Friendly Product With Ambitions

Fractal’s backing as a prominent AI company and its strong relationships with Fortune 500 enterprises position Flyfish for success. The Flyfish team consists of approximately 30 professionals with expertise in AI engineering, product management, experience design, marketing, and research and development. They also have researchers and developers in AI and ML and deep learning.

Flyfish has great ambitions and wants to carve out a niche for itself.

“What we are trying to do is create a global product out of India and that too at the cutting edge of the technology” Marri said.

While there are other platforms like Sendbird which provide an end-to-end automated chatbot experience, Flyfish is looking to differentiate itself from the competition.

Marri also believes that other competitors might take a while to reach the stage where they currently are, and believes that they have made significant progress in innovation and are ahead of the curve.

“At this point of time, there is no direct competition, but there are some people trying to, use chatbots and then bring in a little bit of conventional conversational commerce into the play.

But we’re trying to defy the conventional chatbot experience and bring in a consultative experience.”

Marri revealed that there are several ongoing conversations with the Fortune 500 companies within Fractal’s ecosystem. “That’s one of the biggest advantages that we have, is to use fractals access to large enterprises around the world and then offer flyfish as a product,” he said.

Enterprises have been wary of their data, however, Flyfish prioritises the security of customer data by ensuring that they do not store any of it. Instead, the data flows directly into the enterprise’s ecosystem, including their tech stack, cloud services, CRM systems, and ERP systems, and e-commerce platforms. Moreover, anything that passes through the platform is encrypted for additional measures to safeguard the data.

The Existing AI in Sales Market

While, Fractal claims that Flyfish is the world’s first 360-degree generative AI platform for digital sales. There are other firms that offer several kinds of assistance in sales and marketing.

There are several AI tools available for sales organisations of different sizes. Some notable examples include, Conversica, which provides an automated AI sales assistant that engages leads in conversation and qualifies them before they talk to a sales representative. Exceed.ai, which uses AI to engage in human-like conversations with sales leads via email and chat. Crayon, an AI-powered competitive intelligence tool that tracks competitors’ activities online and automatically generates sales battle cards; and Salesforce, a prominent player in the AI for sales space, with its AI named “Einstein.” Salesforce’s Einstein prioritises leads, evaluates deal likelihood, and allows developers to incorporate AI into Salesforce apps. Microsoft’s Viva Sales offers features such as personalised customer email drafting, customer insights, and generating recommendations and reminders.

The integration of AI into existing platforms by established companies demonstrates the growing trend of incorporating AI into sales processes. According to McKinsey analysts in Harvard Business Review, the capability of AI to enhance and boost sales performance can potentially generate a value ranging from $1.4 to $2.6 trillion in marketing and sales.

The post Flyfish Wants to be a Premium Indian ‘Consultative Sales AI’ Offering appeared first on Analytics India Magazine.

Рубрика: AI