Blog

A new version of Bard is giving tough competition to ChatGPT. Some features introduced are not available on ChatGPT. Bard is available in EU countries and Brazil too. Interaction with Bard is possible in 40 languages. We can upload images on Bard. Bard has the capability to readout the answers. It can store conversation history. These conversations can be pinned for easy retrieval. These could be shared also through links. There is ‘modify response’ feature now. Code could be directly exported to Replit. These are groundbreaking techniques.

LLMs are likely to be used in the field of biology, by learning the language of biology. In previous century, there was considerable research in molecular biology, biochemistry and genetics. It is obvious that biology is programmable. It is decipherable too. There four basic components of life — adenine(A), cystosine (C), guanine (G) and thymine (T). Computers depend on the binary system of 0s and 1s. Biology depends on the quartenary system of A, C, G and T. Here there is conceptual overlap. Proteins are made of amino acids — could be a few dozen amino acids to several thousands of them. There are 20 amino acids to choose from. Thus this too is amenable to computerization.

Denis Hassabis, DeepMind, treats biology as an information processing system. Physics depend on maths as its primary language. AI could as well depend on biology as its primary language.

LLMs work optimally in presence of massive signal-rich data. LLMs infer patents and structures. They generate novel output by comprehending the topic.

By ingesting the whole Internet, ChatGPT has become conversational.

If LLMs are trained on biological data, they could learn the language of life.

In early application, they could be used to design proteins, the building blocks of life. Proteins have shapes as their functions. Antibody proteins target foreign bodies, the antigens, just as the key fits into a lock. Enzymes accelerate biological reactions. They are proteins which bind with certain molecules. This alignment makes us aware about how the life functions.

Protein’s one-dimensional structure was converted to 3D using protein alignment. This was done by using AlphaFold AI system. Of course AlphaFold has not been developed using LLMs. It used MSA — multiple sequence alignment from bioinformatics. But this has limitations. It is a slow compute intensive system. It cannot be used for ‘orphan’ proteins with no analogues. Such proteins constitute 20 per cent of all proteins. Protein structure can be deducted/predicted using LLMs.

LLMs can be trained on protein sequences, instead of the English language. They can efficiently be used to predict the protein structure. It started in 2019. In 2022, Facebook put forward ESM-2 and ESMFold, two powerful protein models. They had 15 billion parameters. The predicted sequences can be reversed, thus paving way to generate novel protein structures.

AI can be used to invent new proteins. The vast unchartered protein space can be explored. It iss a nascent field.

LLMs can be used to generate biomolecules such as nucleic acids.

The ultimate aim is to go beyond modelling. We have to study the interactions of proteins with other molecules, and cells, and tissues, and organs so as to cover the whole living organism.

20th Century was dominated by Physics. It is expected that 21st century will be dominated by Biology.

July 2023. Seven leading IT firms — Alphabet (Google), Amazon, Apple, Meta (Facebook), Microsoft, Nvidia and Tesla — are meeting President Biden to reach an agreement about new AI systems for being transparent and secure.

The AI systems will be subjected to testing, both internal and external, before their release. The systems will be probed for security flaws and discriminatory tendencies.

These firms will make new commitments to share information to improve risk mitigation with governments, civil society and academics. They will report vulnerabilities as they emerge. Leading AI companies will incorporate watermarks into the material they generate. However, this system is yet to be developed. It may prove difficult to stamp content in a way that it cannot be effaced by malignant actors.

There is immense public interest in the emerging technologies, and concern over the societal risks.

The meeting will be important first step in ensuring responsible guardrails for AI.

AI is expected to benefit the society at large, and therefore should be built and deployed responsibly.

Before the present meeting, the executives have already met the VP Kamala Harris.

In July, 2023, Facebook released its open source artificial intelligence model Llama as an alternative to expensive proprietary models of OpenAI and Google. The new version of the AI model is called Llama 2. Its distribution will be done by Microsoft through Azure cloud service. The version will run on Windows operating system.

Previously, the version was available to only select academic institutes for research. It will be made available through direct download and through Amazon Web Services (AWS) and Hugging Face and other service providers.

Opensource provides opportunities to other developers to innovate.

The new version has been trained on 40% more data than its previous counterpart. It has 1 million plus annotations by humans to fine tune its output. Any incremental improvement on open source model eats up the market share of closed source models.

Microsoft wants to give an opportunity to developers to choose a model of their choice. It wants to become a go-to cloud platform for AI.

GPT-3.5 was revolutionary. It collected many users within a short time. Some five months later, GPT-4 was launched (say in March, 2023). It is now time to have a new version of GPT, but OpenAI has just introduced Code Interpreter now. It is a very useful feature, but indicates the company is just adding incremental features. Or is it that the company has already introduced GPT-4.5 with code Interpreter, without officially declaring so. This is just speculation, nothing official so far. Some say that the next version would be super intelligent AI. Altman expected the journey to superintelligence to be of four years.

ChatGPT can be used to visualize data and make diagrams. ChatGPT has introduced code Interpreter. You have to enable it. You can ask data visualisation abilities. It can create line graphs, bar diagrams, pie charts, scatter plots, histograms, heat maps and so on. You have to upload the data in XLS, XLSX and other file formats. You can ask then the code interpreter to ‘make the graph’. You can as well ask the code interpreter to find insights from the data, and visualize it.

If you have subscribed to ChatGPT Plus, you can enable ChatGPT Plug ins to visualize data. You can create diagrams by using Mermaid language syntax. It generate code which is posted on diagram app for visual output.

LLMs, as we know, are good for natural language processing including translations. There are now thousands of models being open sourced on platforms such as GitHub or Hugging Face. Every week about 5000 new translation models are being added.

Big Translate, an LLM developed by a team of Chinese researchers supports multi-lingual translations across 100 languages and it is available on GitHub.

Big Translate is built upon LLaMA of Facebook (introduced in February, 2023). It is designed to handle translation of low-resource language with high accuracy. It is focused on Chinese, and has parallel dataset of 102 languages. The corpus is drawn from various public and proprietary resources.

The model has been tested against Google Translate and ChatGPT. It surpassed ChatGPT in BLEU scores. It closely matches Google Translate.

It can translate Tibetan and Mongolian language. That makes it saleable in the Chinese market. Alibaba Group has released POLYLM to compete with this product.

Advertisers have to choose whether a brand is to be established by telling a drawn-out story or a quick story before people tune out. Today one has to make an impact on the viewer’s mind within 15 seconds. There is a growing trend of shorter ads. 15-secorders are the 30-secorders of today. The attention spans are falling and this is the ideal duration for most of the ads.

Such ads are mostly funny and eye-catching. Consumers switch channels when ads are shown. On digital, it is easier to skip the ad. It is better to have non-skippable ads of 15 seconds.

The impact is created by personalisation, relevance and emotional appeal. There is a hook. The message must bring out the USP. The viewers can be engaged by giving a creative twist.

It is to be noted that ultimately a creative ad engages the audiences in any format.

Almost after a decade of the appearance of online digital assistants such as Siri and Alexa, a new wave of digital assistants could be seen — they are powered by the generative AI technology. And they have greater autonomy.

Silicon Valley wants to leverage the advances in AI and systems powered by generative AI are attracting billions of dollars of investment.

The new digital assistants are called agents or co-pilots. They promise to perform complex personal and professional tasks when directed by the humans. And they do not need close supervision. They act as personal AI friends.

There is a rush to leverage the technology behind foundation models. Individuals, startups and Big Tech –all are in the race.

There is a possibility of human biases sneaking in, and the potential for misinformation. Some fear murderous HAL 9000 from ‘2001 : A Space Odyssey’.

Twitter decided to limit the number of views on posts. It could affect the advertisers. Traditionally, Twitter provided equal opportunities for all advertisers’ followers to see their tweets. The recent change of limiting the views may decline visibility. Consequently, retweets and replies could also drop.

In India ad spending on social media platforms is $1.28 billion (less than 2% of the digital ad spends).

Brands would like stability of the social media platform. Policy changes do affect advertisers. Twitter restricts reading to 1000 posts a day for Twitter Blue accounts, and to 10000 posts for Verified accounts. It harms the bottom line of advertisers. The organic reach is reduced. That means lower brand awareness and engagement. There could be competition for advertising space. It could increase the cost of advertising. Precise targeting will be needed in order to avoid wasting impressions on users who are less likely to engage. Brands must have talents to create content that could maximise impact despite the limited views. The campaign will have to be monitored closely. Engagement rate, click throughs and conversions will matter much more.