AI Credibility Concerns

Plus, 💰 OpenAI Secures Major Funding from SoftBank

Hola Decoder😎

If someone forwarded this to you and you want to Decode the power of AI and be limitless, then subscribe now and Join Decode alongside 30k+ code-breakers untangling AI 🧠.

AI

🤖 New AI Models Raising Concerns About Reliability
Insights from Futurism and Euro News

As artificial intelligence (AI) continues to evolve, new research has uncovered a troubling trend: the most advanced large language models (LLMs) are increasingly likely to give incorrect answers rather than admit they don’t know. While these models excel in handling complex questions, they often fail at basic tasks, raising concerns about their reliability and transparency.

Image by Getty Images

The Decode:

  • A study published in Nature tested several AI models, including OpenAI’s GPT, Meta’s LLaMA, and BigScience’s BLOOM. Researchers from the Universitat Politècnica de València in Spain evaluated the accuracy of these models across various topics like math, science, and geography. They categorized responses as correct, incorrect, or avoidant, where the model admits it lacks sufficient information to answer.

  • The findings revealed that while newer models handled complex problems better, they were also more likely to generate wrong answers than earlier versions. Older LLMs, like GPT-3.5, would often provide avoidant answers, indicating uncertainty. In contrast, models like GPT-4 and BLOOM displayed a significant drop in such responses, instead guessing and offering false information even for simpler questions.

  • According to the study’s coauthor, JosĂŠ HernĂĄndez-Orallo, AI systems “are getting better at pretending to be knowledgeable,” leading to a rise in both correct and incorrect answers. The research found that as these LLMs increase in sophistication, so does their tendency to “make up” information, especially when faced with challenging questions. This creates a paradox where models appear more advanced but also more prone to error.

  • Interestingly, the study also found that humans struggle to evaluate AI-generated answers accurately. Participants misjudged the correctness of chatbot responses between 10% to 40% of the time, suggesting that users are often impressed by the AI’s handling of complex tasks and may overlook obvious flaws in basic answers.

As AI models become more advanced, their growing tendency to generate false answers raises concerns about their reliability. While they excel in solving complex problems, their reluctance to admit uncertainty can mislead users. Programming these models to acknowledge their limitations could improve their overall trustworthiness and ensure they remain valuable tools in AI-driven applications.

TOGETHER WITH SMARTPROXY

eCommerce Maturity Index ‘24 report

Smartproxy, one of the leading proxy and web scraper infrastructure providers, has just released a free industry-first eCommerce Maturity Index ‘24 report. It analyzes over 245K data points in key eCommerce areas, offering valuable insights on crucial topics like user experience and dynamic pricing trends that allow you to optimize your eCommerce store.

Since data-gathering infrastructure is essential for mature eCommerce businesses, Smartproxy is offering you an exclusive 50% discount SCALEUP50 on all plans. Don’t miss this chance with a best-value provider by claiming your discount here.

OPENAI

💰 OpenAI Secures Major Funding from SoftBank as Apple Exits
Insights from The Information

OpenAI, the maker of ChatGPT, has successfully secured billions of dollars in funding from major investors, including SoftBank, Microsoft, and Thrive Capital. However, Apple has reportedly decided to step back from the funding round, marking a significant shift in OpenAI’s investor lineup.

The Decode:

OpenAI is reportedly raising to $6.5 billion via convertible notes, valuing the company at a staggering $150 billion. This latest funding round sees major commitments from key players:

  • SoftBank has joined the investment, providing a significant financial boost to OpenAI’s war chest.

  • Microsoft is planning to contribute an additional $1 billion, adding to its previous $13 billion investment.

  • Thrive Capital has committed $1 billion with an option to increase its investment by another $1 billion next year, contingent on OpenAI meeting certain revenue targets.

Despite past partnerships and collaborations with OpenAI, Apple is no longer participating in this round, according to reports from the Wall Street Journal. This exit is notable, given Apple’s involvement in AI advancements, including its work with OpenAI technologies in Apple Intelligence.

The funding round coincides with OpenAI’s transition to a for-profit entity, a move that has sparked debate. CEO Sam Altman has denied speculation that he will receive equity as part of the restructuring, despite ongoing rumors.

OpenAI’s ability to secure funding from tech giants and investment firms, even amid restructuring and controversies, highlights its continued dominance in the AI space. With a valuation of $150 billion, it remains a key player in the rapidly expanding AI industry, attracting interest from major investors despite its complex business model. 

MICROSOFT COPILOT PRO

🚀 Microsoft’s New Copilot Takes on ChatGPT
Insights from Windows Latest

Microsoft is gearing up to compete head-to-head with ChatGPT by rolling out a significantly revamped version of its Copilot AI. The upcoming updates include a sleek new interface, faster performance, and additional features that make it more versatile for everyday users.

The Decode:

  • Microsoft’s Copilot is set to introduce a new card-based user interface, which is visually cleaner and optimized for web performance. With the new design, Copilot’s answers will now be delivered faster and more accurately, rivaling ChatGPT. The update comes with an emphasis on personalization, where users can sign in to their Microsoft account, set preferences, and even choose from different voice modes.

  • Copilot’s voice mode offers four voice options—Meadow, Wave, Grove, and Canyon—each catering to different moods or tones, similar to the features offered in ChatGPT Plus.

  • Users can expect smoother interactions as they explore various cards, such as tips for falling asleep or curated suggestions based on AI-generated content. The new design will also include customizable themes, “Day” and “Night,” for an enhanced user experience.

  • Despite the visual overhaul and functional improvements, Copilot still has limitations. For example, users cannot yet upload documents like PDFs, and the newer models like GPT-4o or GPT-o1 are not fully accessible. Some regions have also reported performance issues with the mobile app, which is expected to be addressed in the upcoming update.

  • Microsoft’s Copilot Daily, a new feature in the update, presents curated news content in podcast form. Using AI, the feature allows users to listen to news summaries tailored to their interests, further enhancing the versatility of Copilot as a daily productivity tool.

What’s Coming Next?

The upcoming updates also hint at the arrival of GPT-o1 and improvements in plugin support, which will be rebuilt around the concept of “extensions.” These additions aim to bring Copilot closer to the full potential of models like ChatGPT-4 and enhance its overall functionality.

TOOLS YOU CANNOT MISS

🔍 Smartproxy – Supercharge your data projects with Smartproxy’s affordable, user-friendly solutions. Bypass restrictions and capture the data you need. Use Code SCALEUP50 for 50% off on all plans. 

🍽️ Cheffie – Create personalized recipes and analyze restaurant menus for dietary needs. Eat healthy, stress-free, and deliciously with ease.

🎯 Y-Pod – Plan your life goals with AI-powered insights. Stay organized, motivated, and bridge the gap between dreams and daily actions.

🎨 Polymet – Design user interfaces and prototypes with no experience needed. Just describe your idea, and Polymet generates ready-to-use code.

HOT NEWS

🚀 Cerebras Moves to IPO to Compete with Nvidia

AI chipmaker Cerebras Systems filed for an IPO on Monday, aiming to trade under “CBRS” on Nasdaq. Competing with Nvidia, Cerebras’ WSE-3 chip boasts more cores and memory than Nvidia’s H100. The company posted $136.4M in sales with a $66.6M loss in early 2024. UAE-based G42 committed $1.43B in orders. CEO Andrew Feldman stated, “Our chips push AI boundaries further.”  

🤖 Artisan Secures $11.5M to Launch AI Sales 'Employees'

Artisan, founded in 2023, has secured $11.5 million in seed funding to introduce AI-driven virtual employees for sales teams. Supported by Oliver Jung, Y Combinator, and HubSpot Ventures, the funds will enhance its AI assistant, Ava, which automates lead generation and personalized outreach. Ava utilizes a database of 300 million B2B profiles, optimizing sales processes with Future plans including expansion into marketing and customer success. 

🤖 ByteDance to Develop AI Model Using Huawei Chips

ByteDance is reportedly working on a new AI model using Huawei's Ascend 910B chips to lessen its dependence on U.S. suppliers in light of trade restrictions. The model will focus on AI training and inference, utilizing Huawei chips for less demanding tasks. Despite supply challenges, ByteDance aims to strengthen its AI capabilities and diversify chip sources to stay competitive in the AI space.  

“GUARANTEED” The only AI Tool kit you’ll ever need 🤩

We know that you're a mastermind in your circle, so how about sharing the Decode with your crew?

Share our newsletter with your network, and unlock exclusive access to Decode’s ultimate AI tool book, with 1000+ verified, vetted, and powerful tools across 50+ categories.

Just 2 referrals and you can have it all.

PS: This is not just another toolbook. It’s a handpicked compilation of the best of the best tools that actually do what they claim. Also, the list is updated with new tools every month! So get your hands on the only AI Toolbook you will ever need!

Click to copy & paste your referral link to others:

https://decode.beehiiv.com/subscribe?ref=PLACEHOLDER

Tip: Ask your friends to enter their email and then click “confirm subscription” in their inbox.

You currently have 0 points of 2 points to get The Only AI Toolkit you will ever need😎

Thanks for Decoding with us🥳 

Your feedback is the key to our code! Help us elevate your Decode experience - hit reply and share your input on our content and style.

Keep deciphering the AI enigma, and we'll be back Tomorrow with more coded mysteries unraveled just for you!