David Albrecht 🔮

Chatbots were step 1. AI agents are step 10. And they’re already delivering jaw-dropping results across 46+ companies. Let me break it down. Here's what makes them game-changers: 1. Autonomy: They design their own workflows 2. Adaptability: They learn from each interaction 3. Integration: They work seamlessly with existing systems 4. Problem-solving: They tackle complex, multi-step tasks Real-world impact: • 50% increase in first-contact resolution • 30% reduction in operational costs • 90% customer satisfaction improvement The kicker? This is just the beginning. How do you see AI agents transforming your industry? Follow Ghazi for more.

29

23 Comments

My first conversation with new hires: We're excited to have you. There were other good candidates but you were the best. Some day, however, you are going to leave. If you find a better opportunity somewhere else, there's no hard feelings from me. Hell, recruit me to there. But while we take a paycheck, we owe 100% of our effort to making this the best place it can be for each of us. I'm going to be pissed if you leave over something I could have fixed. Don't check out. I can't promise I'll always be able to get you the work and experiences you are looking for, but I'll do whatever I can. Talk to me. If you have issues, I need to know. If your issues are with me, bring them to me anyway. Or talk to my boss, or the head of HR, or the CEO. You will know what you are supposed to be working on and have clear line of sight from that work to the benefit for the business and our customers. Hold me accountable for that. If you lose sight, let me know and we'll figure out if you just need some more information or if we need to change direction. You always need to be able to answer these three questions: 1) What are you working on? 2) Why are you working on it? 3) How will you know when you are done? What do you need from me?

634

35 Comments

Fascinating to see the salary range for Machine Learning Engineers surpassing that of Data Scientists . This shift suggests a growing emphasis on operationalizing AI and ML models rather than just research. Are businesses finally moving from AI hype to practical implementation? What's your experience? #MachineLearning #SoftwareEngineer #AI #Salarytrends #Hiring - Mitchell Martin Inc. Tarah Healy Michael Fromowitz Viola Sarkantyus Read the full report here: https://lnkd.in/ebsD2ikj

13

I've noticed more companies stepping back from building RAG systems. Reconstructing a decades-old discipline like information retrieval isn't easy, even with advanced LLMs. Additionally, evaluating and monitoring RAG pipelines remains challenging and unresolved. An often overlooked point: you don't necessarily need vector search to build a #RAG. Traditional methods like #BM25 can be more effective, consistently performant, and much cheaper. While retrieving certain facts from a collection of documents is fairly straightforward, the most ambitious promise of RAG—answering open-ended questions grounded in a large body of multimodal content—remains unfulfilled. This is challenging because relevant documents might not be semantically close to the question in the embedded space. A potential solution could be using LLMs' internal knowledge to "seed" various searches, combine the results, and then use LLM reasoning to sift through a large context to refine and answer the question. Is anyone exploring this approach?

24

1 Comment

Talking about AI taking your job: “It’s designed to take on a recruiter’s most repetitive task so they can spend more time on the most impactful part of their jobs” or is it AI giving you a job.... https://lnkd.in/eZwdQaTQ #generativeAI #nlp #artificialintelligence

1

I wonder what the click rate is for those LinkedIn AI suggestions on the bottom of posts. Corollary: I wonder what percentage of those clicks are accidental and whether it is three digits.

3

5 Comments

I'm looking for better options to NER on .html files. I'm currently using BeautifulSoup (lxml parser)-> stanza (which seems to perform better than Spacy), but I'm getting a lot of junk classified as an entity. I'm trying desperately to get better off the shelf results with an understanding that there will be a post extraction filtering phase to further refine. thanks! #nlp ##datascience #ai

9

10 Comments

There was a time when AI agents could be built with a basic RAG architecture alongside prompt optimizing fully managed, closed-source commercial model endpoints. The bar to entry was lower, and folks were even looked down upon for simply building wrappers. I never bought into that mindset -- as we all know, solving user problems matters more than the underlying tech. However, in just 20 months of mass adoption of LLMs, we have started to see insane optimization of the entire stack, from a mixture of custom-trained models and embeddings per task to novel re-rankers. Honeycomb and Cursor are the latest examples of setting the bar really high in this space. It's time for AI practitioners like us also to raise the learning and innovation bar for ourselves. Personally, I'm fortunate to be surrounded by highly talented and driven folks who push the boundaries of what a small team can do to make AI systems production-ready. As the field matures, it's clear that the future belongs to those who can rapidly optimize these technologies to create meaningful, impactful solutions. #ai #genai #llm #deeplearning #growthmindset #career #hiring #ml #machinelearning

32

2 Comments

AI is supposed to be dangerous but I've yet to hear a reason that isn't already a problem without AI. So AI lets you fake a video. Well, we've all seen the Biden "phantom chair" squatting video that didn't require AI to be completely misleading. AI is hoovering up all our data and spying on us. Social media and digital ads beat it there. Now I'm seeing hand-wringing about AI papers destroying the peer review publication system. In each case, AI is just exposing the problems that were already there, while elevating the attention paid. And that's why I'm bullish on the future. The answer to all of these frauds is a better social immune system powered by critical thinking, healthy skepticism, and the ability to recognize that just because new information confirms our opinions that doesn't make it accurate. I don't know why it took AI for people to stop trusting what they see on the internet, but I'm good with whatever works.

53

33 Comments

Here's why OpenAI o1 is a big deal. Up until now, the only way to make a model smarter was through training. You had to train on more data with more compute. These training runs are expensive. ChatGPT 4 cost about $100 million to train. Future models are expected to cost $1bn to train, then $10bn. You get where this is going. Once the model is trained, it's pretty much as smart as it's going to get. What they've done with OpenAI o1 is develop a technique where it can get smarter at run time (inference) just by pondering the problem longer. How much smarter it gets scales with how long it thinks. This is a breakthrough.

21

8 Comments

Love the parallels of recruiter mental state to the "hungry judge effect," where decision-making quality can be influenced by factors like hunger, mood, or time of day. Studies have shown that judges are more likely to issue harsher rulings before meals, a phenomenon that underscores how human judgment can be inconsistent and biased by external factors. AI interviewers, on the other hand, don’t suffer from these human fallibilities—they’re consistent, fair, and unbiased. They work 24/7 and they never get tired of the same old questions so recruiters can focus on closing candidates and hiring. People are often skeptical of AI recruiting at first, but that skepticism tends to fade once they see the results. We recently had a recruiter who was initially hesitant to adopt AI for interviews. After just two weeks, they discovered that they could tweak the interview scripts to better match their needs and scale their recruiting output by 10x asynchronously. The AI became not just a tool but a trusted partner, freeing up their time to focus on strategic tasks instead of screening candidates (and go on vacation without worrying about leaving candidates on the hook). It’s not about replacing the recruiter—it’s about supercharging their capabilities and eliminating inconsistencies that can hold them back.

7

Practical vertical uses of LLMs are happening now. The menial parts of 6-figure jobs are being automated away. If you aren’t getting 100% reliability you aren’t chopping down the prompts enough. Don’t rawdog your prompts: write evals and treat it like test driven dev.

1,043

66 Comments

Verticalized AI companies that are building for legacy industries such as automotive, manufacturing and construction have a big opportunity to build new solutions that could have not been possible before. We see an opportunity here and companies like Harvey & Trunk Tools are building around this opportunity as well

9

2 Comments

Something feels very off with Anthropic's latest model, Claude 3.5 Haiku, which you can see in the attached (data from Artificial Analysis). Anthropic says that, since the model is smarter than their best model from earlier this year, "we've increased pricing for Claude 3.5 Haiku to reflect its increase in intelligence." But if you look at comparable options, like Gemini 1.5 Flash and GPT-4o-mini, Claude 3.5 Haiku has worse quality, slower speed, and a much higher cost. This goes against long-term trends. Models have been getting smarter, faster, and cheaper in combination. Today's GPT-4o-mini is, for example, about 100X cheaper than GPT-3 (per thousand output tokens), while being faster and smarter. Anthropic seems to claim that it's worth the price because Claude 3.5 Haiku is the best of all the small models at coding. But is that really how most people will use Haiku? For coding assistance, most people will use a top model like Claude 3.5 Sonnet. Smaller models tend to be for larger scale tasks where the cost of tokens can add up. So what might really be going on? There have been rumors that Anthropic had a failed training run for Claude 3.5 Opus, which people were also expecting along with the release of 3.5 Haiku. There have also been rumors that they're behind on creating a reasoning model to compete against OpenAI's o1. And there are rumors they're raising more money. All of these pressures may have forced their hand here. It's speculation, of course, but it's hard to understand why they would do this if they weren't under such pressures.

9

While going for a run yesterday and listening to an AI-curated playlist on Spotify, I thought, "All this new music sounds the same." 🤔 It turns out I was right! AI tools, modeling software, and drum libraries have converged the music creation process to the point where people with no musical skills can create hit songs. This has been detrimental to creativity and the music creation process. 😥 As a musician 🎸 and AI practitioner 🤖, I am sometimes asked about the intersection of the two. This quote sums it up nicely. Creative AI tools can be seen as sophisticated plagiarism software, as they do not produce genuinely original content but rather emulate and modify existing works by artists subtly enough to circumvent copyright law. In other words, AI can help non-creative people recycle old ideas while inspiring original content, which requires creative mojo. ✨ Check out this video by Rick Beato where he walks through why music has become homogenous and how AI and other tools helped us get here. 📺 It's time for regulation to compensate artists, writers, and content creators who have provided all the data to train companies' profit-making models. #Ai #machinelearning #music #aimusic #regulation #creativity #aiart #spotify

25

5 Comments

I would add three more Vahid Jozi, but they could be "derivatives" but worthy to include as they take a special angle where product value props should focus more on the person and not the company (or the value the company gets from your product because people buy not companies)... Emotional Appeal: Many successful brands build value through emotional resonance, offering comfort, trust, or inspiration. Think of products like Apple or Nike—beyond function, they sell a feeling, a community, or a lifestyle. Sense of Purpose: Increasingly, customers look for brands that align with their values, such as sustainability or social justice. This goes beyond social status; it’s about feeling part of something bigger. Reducing Stress or Providing Comfort: Stress relief and well-being are powerful motivators. Products like wellness apps, cozy home items, or even insurance play into our need for comfort and peace of mind, which aren’t always easily categorized as just saving time or money.

1

"Gen AI video" is so damn bad. Seriously, it's laughably bad. There is no acting, no cinematography, no directing, no pacing, no art direction, no production design, no visual design, no style, just hallucinations and ugliness. Watch the teeth, they change every time the mouth closes and reopens. For all the AI enthusiasts and AI evangelists? This won't get much better because... it is missing HUMAN ITERATIVE CREATIVE INPUT. It turns out that is magic. Who would have thought?!?! Enjoy the nonsense. GEN AI will never draw money with consumers. Its place is AI slop output for social media. Just like this. This is the perfect use of Gen AI. Social media comedy. #ai #fakeai #genai #degenai #generativeai #degenartiveai #aivideo #aislop #aiart #aishitoutput

77

29 Comments

We’ve all heard about RAG (Retrieval Augmented Generation), but now there’s something even more exciting—RIG (Retrieval Interleaved Generation) RIG not only taps into external knowledge like RAG, but interleaves that information directly into the text, creating smoother, more accurate and context-rich responses from large language models (LLMs). Why RIG is a game-changer: ✅ Improved Accuracy: Fact-checks in real-time to reduce hallucinations. 💡 Deeper Contextual Understanding: Captures the nuances of queries with ease. 📊 More Informative Responses: Effortlessly integrates relevant knowledge for richer outputs. More Curious how RIG is transforming the way LLMs work... Read more about it here: https://lnkd.in/g3cPC-tW #RIG #AI #LLM #NaturalLanguageProcessing #Innovation #FutureOfAI

21