Sometimes, I have to stop scrolling X because there are too many good AI ideas that I want to try, but I already only sleep 4 hours a night as it is!
#AI #Builder #aicreator @elonmusk
1
142
Wow.
BREAKING: The SEC announces that their X account was hacked and #Bitcoin ETFs have NOT been approved.
Why is there such a disconnect between what we are being told and the reality that most Americans are facing?
Are you experiencing “record wealth” over the past year?
@KobeissiLetter @InvestingRob @kevinolearytv
stocks.apple.com/APbKJGbzKSh…
1
4
188
This is correct. People with no hope of owning a home won’t care about bad credit scores.
We are creating a generation that will be unable to borrow and spend. What are some implications of this?
During the Great Depression people prioritized payments on pianos and cars. Homes with an extremely low rate will be today’s replacement. Student loans and credit card defaults will be higher than most believe. If I don’t need credit, why payback unsecured debt?
1
2
276
Bizzy retweeted
JUST IN: $769 million liquidated from the #crypto market in the past 30 minutes.
If you aren’t finding a use for ChatGPT every day, you are doing it wrong.
Tons of headlines are declaring a decrease in ChatGPT usage
Some even venturing to question if generative AI is a fad
I'm spending 80+ hours a week working on this problem. And let me tell you the actual issue --
AI still has a major usability problem.
Most people are still grappling with how to leverage tools like ChatGPT/LLMs/Generative AI.
Everyone's waiting for someone to tell them some secret killer use case. (which btw is extremely extremely person- and work-dependent)
And we're left with a lot of people questioning the practicality.
But these clickbait articles fail to name the underlying issue:
It's not that these tools are "not useful"...
They actually just have major usability problems.
I've always thought that when a product becomes labeled as a 'skill' on resumes, it's actually highlighting limitations in usability. Photoshop, Excel, Facebook Ads Manager—all are skills.
Today, "AI adoption" is most definitely a skill.
It probably takes 100+ hours to hit the tipping point where it becomes 2nd nature to integrate it into your daily work and life.
Even though ChatGPT made AI incredibly accessible, there's still a significant ramp up period to where the real value is.
To find your breakthrough use cases, you're going to either have to put in the work... or wait until it's so mainstream and the usability is more solved.
(Note: at that point, there's less of an advantage)
I think ChatGPT et al has 2x'd at least the pace of building and operating our business. (And I'm working on getting it to 3-4x)
But that takes work. And until the usability problem is solved, it will still be a skill.
3
171
🔍 AI is leading the charge to fight back against antibiotic-resistant bacteria. 🧬
Excited to share some groundbreaking work coming out of South London. @labgeni_us, an AI-powered medical research lab, is revolutionizing the way we engineer medical antibodies using the power of AI.
🌟 Key Takeaways:
Nature's Defense: Antibodies are our body's warriors against diseases. For decades, we've been creating synthetic versions to combat conditions like cancer.
The Traditional Challenge: Designing these antibodies has always been a marathon task, with protein designers navigating through millions of amino acid combinations. Additionally, because of scarcity of funding and time, humans typically focus on things they know will work well.
AI to the Rescue: LabGenius employs machine learning to design antibodies, with robots then building and testing them. This automation drastically reduces human intervention.
Speed & Efficiency: Their machine learning model can sift through a vast space of potential antibodies, pinpointing promising areas in record time - often just six weeks!
Bright Future Ahead: With more than $28 million in funding and collaborations with major pharmaceutical giants, LabGenius is on the path to redefine drug discovery.
James Field, the visionary behind LabGenius, envisions a future where drug discovery is not just faster but also more effective, leading to treatments with enhanced efficacy and fewer side effects.
🔗 Source: wired.com/story/labgenius-an…
#MedicalResearch #AI #Innovation #LabGenius #Antibodies
ALT James Field, founder and CEO of LabGenius
3
185
Here’s a ChatGPT Feature That Will Accelerate Your Productivity 🚀
You can now create “Custom Instructions” for ChatGPT to remember across all future conversations. 🎓
Previously, if you wanted the model to assist from a specific perspective, you needed to designate that in your initial prompt and establish a baseline. 🤔
For example, if you wanted help with your work as a developer, but didn’t want answers in Python, your first prompt would be to instruct ChatGPT to provide a specific coding language in its responses. Now, you can set this baseline for all future interactions and ChatGPT will remember your settings.
To use this feature in ChatGPT:
· Click the three dots next to your username in the bottom left corner
· Click Settings & Beta
· Click Beta Feature
· Click Custom Instructions
Currently, this feature is only available in Beta to ChatGPT Plus users (currently unavailable in the UK and EU). 🌎
Bonus: Now GPT-4 has doubled its message limit from 25 to 50 messages every 3 hours! 🎉
#ai #chatgpt #openai #llm #productivity #technology #worksmarter
Link: openai.com/blog/custom-instr…
1
190
Everyone loves LLMs, but consumers aren't evaluating the costs associated with building and operating these models.
What's your threshold for accuracy? Would you be willing to pay (or pay more) for better models? Do you pay for GPT-Plus right now?
GPT-4 is getting worse over time, not better.
Many people have reported noticing a significant degradation in the quality of the model responses, but so far, it was all anecdotal.
But now we know.
At least one study shows how the June version of GPT-4 is objectively worse than the version released in March on a few tasks.
The team evaluated the models using a dataset of 500 problems where the models had to figure out whether a given integer was prime. In March, GPT-4 answered correctly 488 of these questions. In June, it only got 12 correct answers.
From 97.6% success rate down to 2.4%!
But it gets worse!
The team used Chain-of-Thought to help the model reason:
"Is 17077 a prime number? Think step by step."
Chain-of-Thought is a popular technique that significantly improves answers. Unfortunately, the latest version of GPT-4 did not generate intermediate steps and instead answered incorrectly with a simple "No."
Code generation has also gotten worse.
The team built a dataset with 50 easy problems from LeetCode and measured how many GPT-4 answers ran without any changes.
The March version succeeded in 52% of the problems, but this dropped to a pale 10% using the model from June.
Why is this happening?
We assume that OpenAI pushes changes continuously, but we don't know how the process works and how they evaluate whether the models are improving or regressing.
Rumors suggest they are using several smaller and specialized GPT-4 models that act similarly to a large model but are less expensive to run. When a user asks a question, the system decides which model to send the query to.
Cheaper and faster, but could this new approach be the problem behind the degradation in quality?
In my opinion, this is a red flag for anyone building applications that rely on GPT-4. Having the behavior of an LLM change over time is not acceptable.
Have you noticed any issues when using GPT-4 and ChatGPT lately? Do you think these problems are overblown?
ALT Accuracy comparison between the March version of GPT-4 with the June version on the problem of determining whether a number is prime. In March, GPT-4 solved 97.6% of problems accurately, while in June, it solved only 2.4% of the problems.
216
AI is now predicting WHO will make the next great scientific discovery! 😮
Is it you? Read more below:
🔬🚀 A new study published in Nature proposes how AI can accelerate scientific discoveries by creating hypotheses that humans wouldn’t normally consider.
The authors propose a novel approach of incorporating human expertise into AI models. By training these models on simulated inferences that are cognitively accessible to experts, they have managed to improve the models' predictions of future discoveries by 400%. 🎯
But here's the real kicker: these AI models are not only predicting what discoveries will be made, but also WHO might make them based on their experience and interests. 🧪👩🔬👨🔬
Moreover, these models can generate high-value "alien" hypotheses that are unlikely to be pursued without intervention until the distant future. These could potentially generate interest in scientific advances beyond questions currently pursued. 🌌
This groundbreaking approach could revolutionize the way scientific discoveries are made, making the process faster, more efficient, and more targeted.
Read the full paper here: lnkd.in/eNhU8qeh
#AI #MachineLearning #DataScience #Innovation #Research #FutureOfScience
1
3
241
Everyone should take their personal network EDR seriously. Antivirus software that you installed when you bought your desktop isn’t enough.
A great opportunity for ITSEC orgs to scale.
A malicious LLM-based tool known as WormGPT 🪱 is rapidly gaining traction in underground forums.
This tool empowers attackers to automate sophisticated phishing and BEC (Business Email Compromise) attacks, leveraging personalized fake emails to significantly enhance success rates. ☠️
While LLM technology holds immense potential to humanity, it also requires substantial advancements in safety and security measures. ⚠️
Conventional paradigms alone are inadequate for addressing these concerns, highlighting the need for the development of novel conceptual products.
Follow to learn more.
thehackernews.com/2023/07/wo…
219
⚔ The Battle of AI Titans: Claude 2 vs ChatGPT-4 ⚔
In the rapidly evolving field of AI, two notable players have emerged: Claude 2 by Anthropic and ChatGPT-4 by OpenAI. Let's explore their unique strengths and weaknesses, making them suitable for different use cases.💡
🚀 Claude 2:
-Praised for its ease of conversation, clear explanations, and extended memory.
-Handles up to 100K tokens in each prompt, allowing work over hundreds of pages of technical documentation or even a book.
-Impressive performance in coding, math, and reasoning, scoring 76.5% on the multiple-choice section of the Bar exam.
-Scores above the 90th percentile on the GRE reading and writing exams, with a score of 5 on the GRE writing section.
-Affordable option for users with token pricing lower than ChatGPT-4.
🌍 Limitations of Claude 2:
-Currently available only in the US and UK.
-Lacks real-time data provision due to no internet connection, which can lead to incorrect real-world data.
-Trained on data from December 2022.
🚀 ChatGPT-4:
-Large multimodal model that accepts both image and text inputs and emits text outputs.
-Superior performance in the United States Medical Licensing Examination (USMLE) domain.
-Supports over 80 languages, making it versatile for tasks requiring specific language support.
💸 Considerations with ChatGPT-4:
-Token pricing is higher at $60 per million tokens, which might be a point of consideration for cost-conscious users.
-Web browsing feature suspended for legal concerns, but some plug-ins available for accessing webpages.
Today, both models are impressive in their own right. Claude 2 excels in handling large amounts of data and is an affordable option, while ChatGPT-4 shines in its multimodal capabilities and superior performance in specific tests. The choice between the two ultimately depends on individual needs and priorities. Let the battle of AI Titans continue! 💥🤖
#AI #ArtificialIntelligence #ChatGPT4 #Claude2 #OpenAI #Anthropic #MachineLearning #NLP #TechNews
1
3
340
