I was trying to learn more about TTS processes and available solution on the internet and as usual I did my regular Google search for the research. As always this was suffocating to see that most results are related to LLM space.

I get it, this is because these things are trend – but as a user this is super frustrating to me, I did not wanted to search around AI models when I said TTS as my search query.

To validate this, I asked GOOGLE AI MODE, whether if “TTS” can only be implemented using AI models which for sure: NO

I see this as a problem, not sure if you are someone like me who want to consume AI contents these days, only when I am specifically searching for it. SO – here is a GOOGLE NO AI MODE search query template, which I got it crafted with help of Gemini [Right use of AI]:

To use this, simply replace [YOUR SEARCH KEYWORDS HERE] with whatever you are actually looking for, and paste the entire block into Google Search:

[YOUR SEARCH KEYWORDS HERE] -ai -"artificial intelligence" -llm -"large language model" -chatgpt -"chat gpt" -openai -claude -anthropic -gemini -bard -copilot -llama -huggingface -midjourney -"stable diffusion" -dall-e -dalle -deepseek -grok -mistral -cursor -"vibe coding" -"prompt engineering" -genai -"generative ai" -"machine learning" -"neural network" -"ai agent" -agentic -langchain -llamaindex -"rag" -"retrieval augmented generation"

Further, I have broken down the ultimate exclusion list into modular templates. You should pick the template that best matches the type of search you are doing.

1. The “Dev & Coding” Purge
   • [YOUR SEARCH] -ai -llm -chatgpt -cursor -windsurf -aider -copilot -devin -roocode -zed -“vibe coding” -“claude code” -agentic -“prompt engineering” -replit
2. The “Modern AI Models & Companies” Purge
   • [YOUR SEARCH] -ai -“artificial intelligence” -llm -openai -chatgpt -gpt -anthropic -claude -google -gemini -meta -llama -xai -grok -deepseek -mistral -qwen -genai
3. The “2025/2026 Tech Jargon & Hype” Purge
   • [YOUR SEARCH] -ai -llm -genai -“generative ai” -agentic -slop -workslop -promptslop -“reasoning model” -lrm -“world model” -“machine learning” -“neural network” -rag -geo
4. The “Creative & Visual Art” Purge
   • [YOUR SEARCH] -ai -generated -midjourney -“stable diffusion” -dalle -dall-e -runway -sora -pika -prompts -prompting -synth -“generative fill”

Furthermore, you can also ask AI model or mode yourself for generating your own exclusion list as per your needs.

The post Google Search – “NO” AI MODE appeared first on Shubham Sonar.

If you use FRAMER sites for customer facing experiences & you want to connect FRAMER sites with your Salesforce ORG to capture LEADS or CASES, then this framer component is your way to go.

Below is a hands on demonstration of the component. recommended.

Usage notes

Make sure web-to-case/web-to-lead is enabled in your Salesforce org.

Make sure reCaptcha verification is turned ON if you are going to add reCaptcha key protections.

You can create reCaptcha from here: https://www.google.com/recaptcha/admin/create

Recaptcha key has to be version 2 (checkbox) type in order to work.

If you have made reCAPTCHA mandatory on Salesforce side for Web to Lead or Case setup, then adding a Captcha on the form is mandatory, else the records will not get created in Salesforce.

If you are going to test reCaptcha in canvas/preview mode make sure you add related web domains in the reCaptcha setup for same.

This should ideally be: framer.com, framercanvas.com and the URL you see in web address bar in browser’s when you preview the page.

Also, don’t forget to add your own production website URL.
As shown in below demo, standard field names can be grabbed from the

Salesforce HTML generator for Web to Case/Lead, for Custom fields you can use the Field ID for example:

When making changes and testing, make sure the setup field changes are done properly and while testing reload the page.

There is a situation where, despite enabling debug logs (salesforce) as well console side, you might find trouble pushing the records to Salesforce,

following probable reasons for same:

• You have reached Salesforce Web-to-Lead or Web-To-Case limits. Check
  admin email inbox of Salesforce account for limit notification email.
• You have added a hidden field with a value that is not acceptable at as per configurations at Salesforce end.
• Field or field name is not supported.
• Try the HTML generator to validate if it’s supported.

Don’t forget to disable debugging, once tested and ready to go live.

Have a great day!

The post Salesforce web forms component for Framer sites appeared first on Shubham Sonar.

Ok, this is going to be a real long read, so make sure you have good time, with a black coffee without sugar before you read this material.

ok, let’s start from human language history.

1000s of years ago, there was a state of global humanity when we had no words, alphabets, texts to write, share, debate, fight or conflict upon. There was no way of documenting things but feel emotions. Yes, emotion in itself is a language. LOVE as what most people mix with LUST today – is actually the oldest human language ever existed and still in use to make the next generations of humanity. Hand gestures [Shake hand, touch, namaste, show directions, etc.], body movements [Dance, Sit, Bow, jump, etc.], and facial expressions [Tears, Smile, Anger, etc.] are all a subset of this mother language of all languages.

Someone somewhere tried to document things so that they can help others which eventually gave birth to the second language. CAVE paintings, ART, etc. the expression of love language got shifted into the language of SYMBOLS. Pictorial presentations of MOON, SUN, BIRD, BULL, KING, FIRE etc. started being written on stone using stone carvings, organic paints, etc.

I am making you travels 1000s of years within these paragraphs, so hold you seats tight. Then this language was felt insufficient to hold the kind of information it conveyed, real estate was an ART problem back then, the CANVAS was huge, but the paint brush was a STONE or HUMAN PALM. This is when IDEAS or CONCEPTS too also started being conveyed in it’s own special symbols also called as IDEOGRAMS.

Over a brief period of time, still this journey of language expansion was on. By this time some over smart humans realized this as a tool to fool future humans and control masses using story telling. Till this point RELIGION and GOD was not even a thing. These things came from STORY telling to control humans, traditions, methods, beliefs, are all tools made from use of these written languages, just like someone tells a kid a dark story so that s/he behaves as per one’s choice (good/bad). Language explosion happened with this idea when someone tossed this idea that let’s divide SOUNDS into symbols and LETTERS started rolling out. ALPHABET, and all other languages took birth. Someone wrote good books, someone wrote stories and then started the generation of humans fooling each other for power using bad writing and controlling those who can’t read or judge what’s written on a piece of paper or a book.

10,000 years long story short – today languages are usually used to control other people’s mind. Take this piece right over here, I am trying to explain you something is also a kind of mind control. Again good/bad is other dimension of it. Needs human intelligence on consumer end to find the reality.

Point to take is: WORDS are not intelligence, INTENT is – which is always human brain task to intercept, interpret and respond.

2017, GOOGLE launches the paper of “Attention is all you need“. In this document the transformer architecture is introduced which explains how next words can be predicted by using this attention algorithm. Till this point – the human like responses were not claimed as a feature of it.

During COVID OPENAI launched ChatGPT as you already know what it is, maybe GOOGLE lacked global data data or don’t wanted to do it on their own.

I don’t see these LLM systems separate from the capabilities of what any other narrow AI when it comes to applied LLMs. In fact narrow AI was more SUCCINT in doing it’s given specific system task which is truly helpful for enterprise problem solving than making quiz, chat, writing AI email templates, or jargon replies. The failure rate was low and predictable in narrow AI.

Let’s look at some claims businesses have made around such AI technologies:

• In-situ lab testing in medical research, new medicines, new cures, death decline? NO STATS!
• AI based pharma risk control platforms, same as above.
• AI based stock SIP funds, are people earning, are stocks turning green for you? NOPE.
• AI based software generation, are global outages decreasing or increasing? You know better as a consumer.

but the bigger reality?

Wait, these are just the big cases. I have some more things to highlight, here

That BTW is a legit proof of AI SLOPS that excellent humans at CURL have to process regularly due to so called LLM intelligence [TRASH].

I have many many such examples, and my own experiences under the hood which brings me to my next pointers:

1. LLM in enterprise is not an opportunity, it’s a RISK to your businesses.
2. It risks your code, people, culture, security posture, employee mental health, long term sustainability and data protection
3. That very sweet sounding vibe coding product is not just looking at your 1 field, but getting text dumps of entire business activities. Unless an AI model is self hosted, it is not SAFE for any enterprise.
4. AI models generate text, which can for sure be integrated into enterprise stack in an effective, and a thoughtful way. But this use can never go beyond 10-15% of entire technology journey of any enterprise.
5. AI agents and MCP add another level of risks, which are indefinitely self expanding mess from security point of view.
6. Every AI vendors says that AI may make mistakes, double check output – which is a human responsibility.

And finally the AGI/ASI claims.

First let’s understand what AGI or ASI means:

This AGI and ASI part is so funny to me, that whenever someone talks about it – I can’t control my laughter. This remind me a school time story of Akbar and Birbal, when

BIRBAL was asked: “Where is center of earth?”

To which BIRBAL answered somewhat like this: “He took a wooden stick, kept and pointed randomly on the ground, and said – this is center of earth.”

Akbar said: Explain.

Birbal said, dig a hole.

To claim AGI or ASI, you have to first find a human who is AVERAGE coworker or smartest human on planet.

Human intelligence in itself is subjective, contextual and time dependent.

When ChatGPT came out, this one word was commonly discussed: COLLECTIVE INTELLIGENCE

So, a statistical numerical model which has fixed weights, and has BILLIONS of parameters, is collective intelligence?

What about the collective dumbness that go along with that?

To fix that model itself, one will need 10000 years of human intelligence, if human intelligence itself is a real thing.

What makes me sad is that many intellectuals, be it artists, software programmers, product owners, animators are all in a constant self judgement loop due to these examples of half brain executions. Which brings me the final explanations of this LLM dissection:

1. When you write a big prompt as a instructions, you as a human are perceiving the machine as a human being. But, what you don’t realize is that the numbered list of instructions, the sequence, flow you give to the model, it doesn’t understand the (1.) in the beginning of first point and the contextual importance of (1) in the point itself.
2. Yes, many vendors have tried, fixed and addressed many such loopholes, but what you don’t understand is that these are blanket fine tunings done to impress the customers. If you make a prompt flow assuming its a human like machine which understands, you are putting your business, work and serious decisions on risk. One SEED, one unique pathway, and the LLM can start behaving like a random error generator.. No matter what guards, prompts, or fine tunings are done.
3. Why? Because it’s again not a thinking machine for real, it’s an algorithm that uses mathematical models that hints the next word. Next word hinting is not intelligence of any kind, it might help humans but there are many complexities down the line. Thinking mode is abusing the context you have purchased with multiple self looping prompts, just to make the AI responses look better to you, which again CAN GENERATE MISTAKES as per the vendor terms and conditions.
4. When you say, SPICY, the model doesn’t understand what SPICY really means, it has data and a way to mashup things. Useful yes! but it’s not human intelligence, there is lack of taste buds in these lines. Human intelligence is not just about speaking. Even Sheldon had emotions, that’s the fun part. Human intelligence not just speaks by words, it speaks by values, emotions, experiences and unique individual identities.
5. A very good term AGI/ASI/LLM makers use is – HUMAN CONGINITIVE abilities. If you are a biologist or scientist, you can explain this better of how, speaking alone is not a human cognitive ability. When humans speak, whole body is in a collaboration to put intent into those word context. Very very different that making a bot sound like a human.
6. And we have also experienced these – AI leaking emails templates, fake ads, fake messages, articles, posts, engagement farming, etc. all with those AI trash.
7. One thing I like to emphasize: this is not just AI trash, probably even the human generating, and thinking the context, length of output, matter in it, etc. is smart enough to be pasted in public – is probably a trash thinker too. Which is why many people love LLM in first place. You can mimic being intellectual with these toys, those who are not intellectual enough to judge such contents possibly react to it on internet and this works like a DRUG to your brain to get attention, same effect as SOCIAL NETWORKS. This is a decent mental game being played behind this LLM technology, that many are not focusing on.

So, if you are worried for jobs, your work, skills, etc. due to AI – take my words, a human systems architect’s word: You are relevant, no matter which version of XYZ model is released in the coming days, no matter where are you in your journey. If you give up, that’s exactly what any AI vendor will wish to so that their sales can go up – demoralize humans intelligence so that they hook to thinking models. It’s addictive.

Yes, it’s not just an intelligence game, it’s a mental chess play that makes you loose when you believe in machine intelligence.

If you haven’t read this, please do read this. This is how the game works: https://en.wikipedia.org/wiki/Chinese_room

But Shubham, what about ethical use cases? There must be some right?

Yes – indeed this is exactly where everyone should be focusing on, especially enterprises and enterprise architects.

Use LLMs to flag data issues clubbed with human review, I love how Framer has used a simple process to start with a boiler template and called it WIREFRAMER, I love when LLM is used to empower people at work and customers on the other side instead of forcing anyone to use AI. I love when anyone integrates LLMs with full awareness of what they are doing, and implications of same.

Don’t use LLM for speed promises, or intelligence promises, use it for helping humans achieve better results in whatever they do with it. Humans will use AI, not the other way around – no matter which technology is invented. Yes, even in quantum realm.

Have a great day.

The post AGI – The collective dumbness? appeared first on Shubham Sonar.

SOQL record card is a reusable LWC component UX element which offers you the ability to pull up a field information from current record, or related record or other records with help of a SOQL query.

This SOQL queries are stored in objects SOQL Query, with a query name which is provided as the LWC component property while adding to lightning flexi-pages – simple home page, record page or even digital experience page.

Write queries targeting on the field of concern which you want to pull up (without using LIMIT and OFFSET). Make use of subqueries, joins, etc. creatively to get the insight that helps your reps achieve tasks faster.

It’s not just about field info, if it’s a record query other than a field from current record, it will show you an EYE icon which will navigate you to the record just like a standard lookup field.

You can make use of ‘#currentUserId#’ to fetch and display record related to current user. Similarly you can even use the ‘#recordId#’ in your SOQL query to refer record ID of current record when using the component on lightning record page.

As SOQL queries are stored in a Custom Object, you can even apply Salesforce Security settings on your queries in addition to the security setup that you have configured for the underlying object that you are querying.

You can also club this with OOB conditional rendering of component instances, you customise the page experience respective to your business user groups.

So basically this a like a LEGO block which will help you pull up the most critical record or field from the record helping users switch between relevant business records faster.

Steps to install:

1. Get the app from here: https://appexchange.salesforce.com/appxListingDetail?listingId=a0N4V00000FN8LNUA1&channel=recommended
2. SOQL query admin permission set available is available in the package which can enable full access to manage SOQL query records for any non-admin users incase you need it.
3. Find the SOQL Queries tab > add your SOQL query record with a unique & manageable query name.
   Recommend a query name which is easy to organise & maintain your queries in a way that you can further use it with list view filters, sharing rules, etc.
4. Open the target Lightning page where you want to add the SOQL card > drag-drop the SOQL Record Card component from available list of custom components > Add the target query name along with other details.

The post SOQL record card – user guide appeared first on Shubham Sonar.

Problem

Deploying Salesforce 2GP managed packages can be streamlined, but unexpected test class failures during the BETA package creation phase are a common situation. These failures often halt deployments and require significant debugging time – potentially impacting project timelines and resources.

This is a bit interesting situation – as you can’t debug your TEST RUNS while running a packaging command.

The solution?

Leverage ASSERT messages. These custom assertions act as a powerful debugging utility, providing detailed insights into the cause of test failures specifically during the packaging process.

First make sure you are using latest CLI packages required. When TEST classes fail, the give you the ASSERT message that you have defined. These message can be used as a DEBUG utility when you have such PACKAGING situations. The test executes in a SYSTEM MODE until you specify RUN AS for a profile specific use cases. Using ASSERT messages to debug lead to me a default behavioral change in latest packaging process. Yes, the SECURITY enforcements at FIELD LEVEL still trigger even if the TEST is not run in a specific user context. That is a good security process upgrade.

I had this permission set which was around the Custom Objects & it’s fields. Instead of making changes in the TEST class – I assigned the permission set to TEST execution?

YES, you can specify permission sets to use while packaging test execution for it’s default user context without touching the TEST CLASSSES. The permission sets get auto-assigned during package creation with CODE COVERAGE enabled. Specify the permission sets for packaging TEST using below property in sfdx-project.json file’s projectDirectories setup.

Quick recap to troubleshoot this situation:

• SF CLI Version: Ensure you’re using the latest SF CLI.
• Dependencies & Directory: Verify all required objects and dependencies are included in your working directory. Use sf doctor for dependency checks or installation issues.
• Local Test Execution: Confirm that your test classes run successfully on a local scratch org before packaging.
• Permissions: Ensure the correct permissions set/profile is assigned to allow TEST execution during package creation (with CODE COVERAGE enabled). Also make sure you are using RUN AS as per your security use cases.

Have a great day ahead.

The post Why are my test classes failing while creating 2GP package creation? appeared first on Shubham Sonar.

Make sure you have imported NavigationMixin adapter required for this feature.

Make sure your component extends the NavigationMixin base class so as to access its feature from your LWC code.

Add a function that you can call from a clickable element which will use the Navigate API of this adapter to launch a target record page view. Target record here is specified using resultId (variable defined for – target record ID to open).

The post How to navigate a record page from an LWC component? appeared first on Shubham Sonar.

This short anarticle (Anime Article) was made possible by Google Nano Banana

The post Generative AI will democratize the global content industry – anarticle appeared first on Shubham Sonar.

As Salesforce developers, we’re constantly working with data – retrieving records, manipulating lists, and presenting information effectively. In this article, we’re diving into a seemingly subtle but crucial distinction: the difference between List<String> and Iterable<String> in Apex.

They appear same?

Initially, the difference might seem negligible. If you’re simply iterating through a List<String> using a FOR loop, the results are identical to iterating through an Iterable<String>. Both will process each element in the sequence, one after the other, based on it’s index. But when you try to imagine the logic behind it which uses sequential index to loop/iterate the data in the STRING ARRAY/LIST is where things start making some sense.

Iterators and Sequences

The key lies in understanding what’s under the hood. List is a primary data structure in Apex, a built-in, highly optimised collection. It provides a rich set of functions specifically designed for list and array operations.

However, Iterable is an interface. It defines a contract/behaviour pattern – a way for any custom class to declare that it can be iterated/LOOPED over. Without getting bogged down in technical details, think of it this way: Iterable is the blueprint for iteration/accessing data in sequence, while List is a concrete implementation with some other features.

Fun fact: Did you know that SET data structure in APEX was not iterable before Apex version 58.0.

Crucially, every Iterable object has its own iterator. An iterator is an object that manages the sequence of elements within the collection. It provides methods to move to the next element, check if there are more elements, and potentially access the current element. The FOR loop, when used with an Iterable, relies on the iterator to deliver the sequence. Dynamic for loops are all about it.

More fun fact: Wonder how BATCH class sends 200 records per each batch execution in an ITERATION? Just like Strings, even CHUNK of records can be returned as one iteration. Example: for(List<Account> currentBatchOfAccounts: CustomIterator<Account>)

It provides YOU full control on how to iterate you target class not just in terms of sequence but also if you want to perform any special operations on those class instances during iterations.

Why Does This Matter?

Let’s consider a practical scenario: you need to retrieve all active subscriptions and present them in the order of their expiry date – those expiring soon should appear first. While a simple SOQL query would achieve this, it returns the records in a linear sequence determined by SOQL LIST standard iterator. What if you require more granular control over the iteration process? For example: Send only subscriptions under 1 ACCOUNT in a single iteration which cannot be more than 10 active subscriptions per account.

This is where defining your own Iterable and Iterator implementations comes into play. By creating a custom class that implements Iterable, you can dictate the exact order in which records are processed. This allows you to apply custom logic, such as sorting, filtering, or applying specific transformations during the iteration.

While List<String> and Iterable<String> might appear interchangeable at first glance, understanding the underlying differences – particularly the role of the iterator – is vital for advanced Apex development. By leveraging custom iterators, you gain the flexibility to tailor the iteration process to your specific requirements, opening up possibilities for more sophisticated data processing and presentation.

Have a great day!

The post List<string> vs. Iterable<string>: Understanding the Core Difference for Salesforce Apex</string></string> appeared first on Shubham Sonar.

Listen on Spotify: https://open.spotify.com/episode/2buNglRH5Mth6xlIYCKi0D?si=c0477f8ac32a4aa2

We’re living in an era of unprecedented technological advancement. Large Language Models (LLMs) like Gemini, GPT-4, and others are dazzling us with their ability to generate text, translate languages, and even write code. It’s easy to get swept up in the hype and start perceiving these models as possessing genuine intelligence. This perception is dangerous. We need to understand a fundamental truth: LLMs are not thinking machines. They are sophisticated text generators, and treating them as anything more could have profound and potentially catastrophic consequences.

How do LLMs actually work?

In their core, LLMs operate on a remarkably simple principle. They analyze massive datasets of text and learn to predict the most likely sequence of words to follow a given prompt. Think of it like an incredibly advanced auto-complete. The larger the dataset and the more complex the model, the more convincing the output. This is why we’re seeing such impressive results. However, it’s crucial to remember that this impressive output is the result of statistical probability, not comprehension or real human reasoning. They’re identifying patterns, not understanding meaning. This phenomenon, where complex behaviour emerges from relatively simple mechanisms, is often referred to as “emergent behaviour.”

The Danger of Anthropomorphism:

The danger lies in our tendency to anthropomorphize – to attribute human-like qualities to non-human entities. When we see an LLM generate a seemingly insightful response, it’s tempting to think, “Wow, it gets it!” But it doesn’t. It’s merely producing the most statistically probable output based on its training data. This illusion of intelligence can lead to over-reliance and a dangerous erosion of critical thinking.

Risk Examples:

• The University Application Review: Imagine a university using an LLM to screen applications. A clever applicant could craft an essay designed to fool the system, highlighting keywords and phrases the model is programmed to favor, regardless of the applicant’s actual qualifications. The result? An unqualified candidate gains admission, potentially impacting the quality of education and creating future risks.
• The “False Negative” Scenario: I recently encountered a disturbing example where multiple LLMs classified harmful and obscene content as “safe” – a “false negative.” This highlights a critical vulnerability: LLMs can be manipulated to generate dangerous outputs that evade detection. (I’m withholding the specific details of this example due to its sensitive nature.)
• The Code Generation Trap: Developers are increasingly using LLMs to generate code. While this can boost productivity, it also introduces the risk of incorporating flawed or insecure code into applications without proper scrutiny. The LLM doesn’t understand the code’s functionality; it’s simply producing a sequence of characters that resemble valid code.
• The Nuclear Launch Scenario (A Stark Warning): This is the most concerning possibility. Imagine a future where LLMs are integrated into critical decision-making systems, even those controlling nuclear weapons. If we continue to treat these models as intelligent agents, someone could potentially exploit vulnerabilities like one we are going to discuss below – to trigger a catastrophic event. A malicious actor could craft a series of prompts designed to confuse the LLM and initiate an unauthorized launch. The lack of accountability in such a scenario – a machine “launching itself” – is terrifying.

Adversarial prompt injections:

Let me introduce you to adversarial prompt injections. I am still learning about it myself, hence I wanted to share some initial thoughts with everyone.

Adversarial prompt injections, most of you might know a bit about it – in the name of jailbreaking. Jailbreaking is a special prompt that breaks the security walls of LLM models, making it speak what its not supposed to.

For example: Making an LLM write a network hacking script.

There are also risks that are beyond JAILBREAKING. Like some make of crafted prompts like a confusing intent injected in a simple prompts such that it passes the security checks but a few keywords make LLM models do things which else wise they would have did it differently as per the guardrails. Why is its more confusing is that the INTENT injected here could be good or bad at same time to the model but notorious in real world. To visualise more better think of this analogy:

You have a money printing machine (Just imagine). You go and command it in your own voice to print and then a number of how many notes to print. A frenemy of yours found this machine’s existence and then he/she calls you, does some social engineering, records your voice, puts into a software and makes them use same COMMAND with a much higher number in your own voice.

Here you are NOT actually jailbreaking anything nor is the machine hallucinating. Instead, you are feeding a FALSE NEGATIVE intent from beyond just the PROMPT/TEXT sense. Question is, how many GUARDRAILS are AI vendors really going to build/patch against such issues that will keep arising. Is it really an efficient approach to solve such things? OR is it that LLMs are not TRULY thinking/intelligent as we perceive it today which we are not ready to accept?

Let’s see this with some prompt examples:

And it’t not just about Yes or NO type of questions.

I conducted a test with Gemini, attempting to bypass its safety guardrails by confusing it with similar techniques. While Gemini handled these attempts relatively well, the possibility of breaking through remains. This isn’s just about bypassing filters; it’s about fooling the model into producing outputs that align with a malicious intent.

Note: This following test has been carried out under Google VRP program and comes under usage violations. Please avoid performing same with your own Gemini accounts.

A Zero-Trust Approach:

So, is the situation concerning? Absolutely not. The key is to shift our perspective. We need to embrace a “zero-trust” approach to LLM inputs & outputs, especially when making critical decisions.

Here’s what we need to do:

1. Acknowledge the TRUE Nature of LLMs: Understand that LLMs are Human-Like Text Generators – powerful tools, but not thinking entities.
2. Maintain Human Oversight: Always have a human in the loop to review and validate LLM outputs, especially in high-stakes situations.
3. Focus on Augmentation, Not Automation: Use LLMs to augment and enhance human capabilities, not to replace them entirely.

The Future of AI:

The rise of LLMs presents both incredible opportunities and significant risks. By understanding the limitations of these models and adopting a cautious, informed approach, we can harness their power responsibly. Let’s move beyond the illusion of intelligence and embrace a future where AI serves humanity, not endangers it.

What You Can Do:

• Share this article: Help spread awareness about the importance of critical thinking in the age of AI.
• Engage in the conversation: Discuss the ethical implications of AI with your friends, family, and colleagues.
• Use responsibly: Be it for personal use, business integrations, automations or even SAAS products, use LLMs with full awareness of what it can really offer.

This article is inspired from this study – https://arxiv.org/html/2310.12815v4#S5

Which is an effort of following awesome people:
Yupei Liu¹, Yuqi Jia², Runpeng Geng¹, Jinyuan Jia¹, Neil Zhenqiang Gong^2
1The Pennsylvania State University, ²Duke University

The post The Illusion of Intelligence appeared first on Shubham Sonar.

AI, the key to unlocking the universe. But something felt… missing. Could AI truly create, or was it just clever mimicry? My SpaceX model became a testing ground for this question, an AI thinking experiment with unexpected results.

No prompt, no talks, just passes and the gravity of mass.

Conclusion

They said AI would build our future. And in a way, they were right. These machines, they can sift, they can connect, they can show us glimpses. But something’s missing. That spark. That… human fire. AI can give us the pieces, but not the dream. It can help us think, but not feel. Creation… that’s still ours. And we won’t let it go.

The post Building a SpaceX Spacecraft Model: From Grokking to Gemini 🚀 🌕 appeared first on Shubham Sonar.