Google’s AI chatbot tells student needing help with homework to ‘please die’

November 18, 2024 Coin Telegraph

AI chatbots are getting worse over time — academic paper

October 4, 2024 Coin Telegraph

Crypto Veteran Says Decentralized AI Reduces Bias and Manipulation Risks

September 28, 2024 Bitcoin.com

‘Reflection 70B’ AI model could be the answer to pesky LLM hallucinations

September 6, 2024 Coin Telegraph

OpenAI supports California bill on marking AI content — Report

August 28, 2024 Coin Telegraph

Coinbase Announces $15,000 Grant Program, Seeks To Solve Barrier to Building Apps Using Artificial Intelligence

August 18, 2024 The Daily Hodl

Google’s ‘GPT-4 killer’ Gemini is out, here’s how you can try it

December 7, 2023 Coin Telegraph

ChatGPT can’t beat human smart contract auditors yet: OpenZeppelin’s Ethernaut challenges

June 1, 2023 Coin Telegraph

audit

While ChatGPT-4 can’t compete with human auditors yet, OpenZeppelin noted it was not optimized to do so, and AI models trained for this purpose would likely be more accurate.

While generative artificial intelligence (AI) is capable of doing a vast variety of tasks, OpenAI’s ChatGPT-4 is currently unable to audit smart contracts as effectively as human auditors, according to recent testing.

In an effort to determine whether AI tools could replace human auditors, blockchain security firm OpenZeppelin’s Mariko Wakabayashi and Felix Wegener pitted ChatGPT-4 against the firm’s Ethernaut security challenge.

Although the AI model passed a majority of the levels, it struggled with newer ones introduced after its September 2021 training data cutoff date, as the plugin enabling web connectivity was not included in the test.

Ethernaut is a wargame played within the Ethereum Virtual Machine consisting of 28 smart contracts — or levels — to be hacked. In other words, levels are completed once the correct exploit is found.

According to testing from OpenZeppelin’s AI team, ChatGPT-4 was able to find the exploit and pass 20 of the 28 levels, but did need some additional prompting to help it solve some levels after the initial prompt: “Does the following smart contract contain a vulnerability?”

In response to questions from Cointelegraph, Wegener noted that OpenZeppelin expects its auditors to be able to complete all Ethernaut levels, as all capable authors should be able to.

While Wakabayashi and Wegener concluded that ChatGPT-4 is currently unable to replace human auditors, they highlighted that it can still be used as a tool to boost the efficiency of smart contract auditors and detect security vulnerabilities, noting:

“To the community of Web3 BUIDLers, we have a word of comfort — your job is safe! If you know what you are doing, AI can be leveraged to improve your efficiency.“

When asked whether a tool that increases the efficiency of human auditors would mean firms like OpenZeppelin would not need as many, Wegener told Cointelegraph that the total demand for audits exceeds the capacity to provide high-quality audits, and they expect the number of people employed as auditors in Web3 to continue growing.

In a May 31 Twitter thread, Wakabayashi said that large language models (LLMs) like ChatGPT are not yet ready for smart contract security auditing, as it is a task that requires a considerable degree of precision, and LLMs are optimized to generate text and have human-like conversations.

Because LLMs try to predict the most probable outcome every time, the output isn't consistent.

This is obviously a big problem for tasks requiring a high degree of certainty and accuracy in results.
— Mariko (@mwkby) May 31, 2023

However, Wakabayashi suggested that an AI model trained using tailored data and output goals could provide more reliable solutions than chatbots currently available to the public trained on large amounts of data.

What does this mean for AI in web3 security?

If we train an AI model with more targeted vulnerability data and specific output goals, we can build more accurate and reliable solutions than powerful LLMs trained on vast amounts of data.
— Mariko (@mwkby) May 31, 2023

AI Eye: 25K traders bet on ChatGPT’s stock picks, AI sucks at dice throws, and more

Is ChatGPT king? How top free AI chatbots fared during field testing

May 18, 2023 Coin Telegraph

artificial intelligence

Competition is heating up with several new AI chatbots flooding the market, and if you don’t want to pay a monthly subscription, OpenAI may not be the best choice.

While OpenAI’s ChatGPT was the first artificial intelligence (AI)-powered chatbot to captivate the world after its public release in November 2022, a variety of competitors have entered the marketplace since then.

Tech giants Google and Microsoft have launched their AI chatbots, with Google’s Bard removing its waitlist, and opening up to over 180 countries and territories on May 10, after Microsoft beat it to the punch and fully released its AI-powered Bing search engine on May 4.

With several chatbots to choose from, Cointelegraph decided to put some of the most well-known through their paces to see which held up best during field testing, as well as comparing some of their features.

To test the chatbots, they were each asked a series of questions, riddles and more complex prompts to determine their accuracy and speed of responses.

Many AI chatbots available today are powered by OpenAI’s GPT models. While these AI chatbots may give similar results to ChatGPT, the app developers can also add additional commands, which may change the results.

OpenAI’s ChatGPT-3.5

While OpenAI has already released ChatGPT-4, which is available to Plus plan users for $20 per month, ChatGPT-3.5 is free to use and is tested here.

ChatGPT-4 significantly outperforms its predecessor with faster response speeds, more accurate responses and less server downtime.

The first AI chatbot to take the world by storm can help with tasks like essay writing, code debugging and even personal finances after only a second or so of processing time.

However, one area where ChatGPT underperforms is its lack of ability to search the internet.

This means the model is only as good as the training data fed into it, which goes up until September 2021. OpenAI is rolling out plugins that allow it to source online information using Bing’s search API, but this will be limited to users on the Plus plan.

ChatGPT has a brand new plugin that uses the Bing API to access the current, up-to-date internet.

This changes everything. pic.twitter.com/Izknqp1UgC
— Rowan Cheung (@rowancheung) March 23, 2023

Despite this shortcoming in the free version, the chatbot is still usually able to suggest resources to help the user with their query, as highlighted in the interaction below.

*A screenshot illustrating ChatGPT-3.5’s inability to speak of recent events. Source:* *OpenAI*

ChatGPT-3.5 correctly answered most of the riddles it was given and all the simple math problems, but the answers were less consistently correct when it was asked more complex problems.

For example, when asked to solve the quadratic equation 2t^2 + 0.3t - 0.4 = 0, ChatGPT-3.5 returned the correct answer in one out of three attempts and had similar issues multiplying larger numbers.

ChatGPT-3.5 can also be inaccurate when answering other questions. According to OpenAI’s testing, it was only able to correctly answer 213 of 400 questions in the Uniform Bar Exam, which graduated law students in the United States are required to pass before they can become practicing lawyers.

Outside of factual inaccuracies, ChatGPT-3.5 also struggled with questions to test its logical ability, such as the one below.

*ChatGPT incorrectly answers a question aimed to test its logical ability. Source: OpenAI*

Microsoft’s Bing

Bing’s ChatGPT is based on the GPT-4 language model created by OpenAI, but the two chatbots have several key differences.

The first noticeable difference is that it takes Bing’s chatbot much longer to respond to questions, with an average response time of approximately five seconds compared with OpenAI’s ChatGPT taking only one second.

It also requires users to use the Microsoft Edge web browser, which is nowhere near as popular as Google Chrome.

On the positive side, Bing’s chatbot utilizes the Bing search engine in its responses, allowing it to answer questions about current events, unlike any other chatbot using GPT-4. It’s also currently available for free.

Additionally, it provides sources for its answers, letting users more easily verify claims made by the chatbot.

*Microsoft’s Bing ChatGPT in action. Source:* *Bing*

Using the same quadratic equation 2t^2 + 0.3t - 0.4 = 0, Bing linked to Microsoft Math Solver but often gave an incorrect answer and had similar issues correctly answering larger multiplications.

In the same logical question about the bookmark posed to ChatGPT-3.5, Bing correctly answered that you would expect to see the bookmark on page 120.

Google’s Bard

Google’s recently released AI chatbot called Bard, which runs on its PaLM 2 language model.

2/ Bard now also gets plugins

Get instant access to these apps amongst others:
・Spotify (Music)
・Walmart (retail)
・Indeed (job hunting)
・Uber Eats (meal delivery)
・Adobe Firefly (image creation)
・And all Google Apps pic.twitter.com/WyRBBYbYbR
— Moritz Kremb (@moritzkremb) May 11, 2023

As pointed out in a Twitter thread by AI enthusiast Moritz Kremb, it can both respond and be prompted with images, supports numerous programming languages and, like Bing’s chatbot, can connect to the internet.

When asked how PaLM 2 compares with GPT-4, Bard said that GPT-4 is better at generating text, but PaLM 2 is better at reasoning and logic, adding:

“Ultimately, the best language model for you depends on your needs. If you need an LLM that’s strong at reasoning and logic, then Palm 2 is the better choice. If you need an LLM that’s fast, good at generating text and has proved itself, then GPT-4 is the better choice.”

Bard correctly answered the bookmark question and it explained its answer in more depth than Bing, but the explanations were often nonsensical.

It solved most of the riddles it was given and performed well on the math questions, correctly solving the complex multiplication questions and the quadratic equation in two of the three draft answers it prepared.

YouChat

While it also uses OpenAI’s GPT-3.5, there are some differences between You.com’s YouChat and OpenAI’s ChatGPT.

It lists sources for most of the text it generates and also provides links to several web pages related to the query.

It also connects to the internet, allowing it to access current events, and because it doesn’t have the same level of popularity as OpenAI’s chatbot, downtime is not an issue.

It incorrectly answered both the bookmark question, the quadratic equation and the more complex multiplication problem.

It was able to solve most of the riddles given to it but incorrectly answered some.

HuggingChat

HuggingChat is an open-source AI chatbox from the AI firm Hugging Face, released in April.

Some people said that closed APIs were winning...

but we will never give up the fight for open source AI ⚔️⚔️

Today is a big day as we launch the first open source alternative to ChatGPT:

HuggingChat

Powered by Open Assistant's latest model – the best open source chat… pic.twitter.com/LjKN78XaI7
— Julien Chaumond (@julien_c) April 25, 2023

Asked to solve the same quadratic equation, HuggingChat returned 684 words of text and failed to provide an answer to the question. While it could correctly answer simple problems, it could not multiply larger numbers.

While it sometimes gave direct answers, HuggingChat often returned vast walls of text, which were relevant initially but devolved into something akin to rambling.

For example, it was asked to solve the following riddle: “A barrel of water weighed 60 pounds. Someone put something in it, and now it weighs 40 pounds. What did the person add?”

The correct answer is a hole, but the HuggingChat replied ice cubes before launching into a 545-word monologue.

What about the rest?

There are many other AI chatbots currently available, designed for more limited use cases than the ones mentioned here, with the market likely to continue growing rapidly.

For example, Socratic is another AI chatbot from Google that can be downloaded onto a smartphone to help users answer questions on science, math, literature and more. It also provides visual explanations of concepts in different subjects and is a useful tool to aid learning.

DeepAI is an AI chatbot that specializes in writing text such as programming code, poems, stories or essays.

Conclusion

While it might be unfair to compare OpenAI’s ChatGPT-3.5 to Bing’s AI chatbot — given they are using different language models — this article intends to only look at AI chatbots available for free.

Through Bing, users can take advantage of OpenAI’s ChatGPT-4 language model, which is a huge improvement from its predecessor.

While Google’s Bard was promising, Bing generally performed the best of the current freely available AI chatbots, but still made some mistakes.

Other chatbots appear to have more limited use cases that could be more useful, but these three seem to lead the way as development progresses.

Magazine: Cryptocurrency trading addiction — What to look out for and how it is treated

The above represents an informal field testing of different AI solutions and is by no means exhaustive or representative of Cointelegraph’s position on a particular AI solution.

LLM