Big AI news this week! Be sure to check out Pinecone for all your Vector DB needs: Join my newsletter for regular AI updates ππΌ Need AI consulting? π My Links π ππ» Subscribe: ππ» Twitter: ππ» Discord: ππ» Patreon: ππ» Instagram: ππ» Threads: ππ» LinkedIn: Media/Sponsorship Inquiries β
Links:
New INSANE AI Chip, GPT4o Voice Update, Claude 3.5 Dominates, SpaceX Double Landing, AI Video Games
By
admin
Do you think this new AI performance is real?
Yes please, testing Claude sonnet in agent setting would be great.
Here's a good question that we can rarely get right: If I isometrically scale an orbital system by a factor that corresponds to doubling its radius, by what factor does the period change? Answer=1, period and frequency do not ever change when an orbital system is scaled isometrically. Most AI currently will make incorrect assumptions about how these two relate zero shot.
I guess I am one of the few test users of chatgpt voice. (No one notified me, it just turned up as an option on the phone app. ). Its nice but really bad at listening and will run on in the wrong direction with no way to interupt. Does not wait for user to finish thought during conversation if you pause or take a breath
I asked for ChatGPT's opinion, some minor editing.
All this is for entertainment purposes, and it is entertaining.
Much of the excitement surrounding the SOHU chip is based purely on "claims and projections" from the manufacturer. While the technical specifications and potential benefits are impressive, it's important to acknowledge that the chip's real-world performance and market impact are still largely SPECULATIVE.
The Manufacturer Claims: The reported performance metrics, such as being 20 times faster and cheaper than Nvidia's H100 GPUs, come directly from the chip's developers, Etchedβ (OODA Loop)β. These "claims" NEED independent verification through broader industry testing and real-world usage.
Is this an OODA bandwagon ?
EARLY HYPE: The positive reception and hype are based on the "theoretical" capabilities of the chip. Media reports and analyses highlight its potential to disrupt the AI hardware market and offer significant advancements in AI applicationsβ (Geeky Gadgets)ββ (MarkTechPost)β.
Market Entry: The true impact of the Sohu chip will depend on its adoption by major companies who real world experience integrating actual real world Nvidia hardware into their existing AI infrastructures. This process can be slow and is often fraught with technical and logistical challenges.
HAVE YOU NEVER HEARD OF VAPORWARE !!!!
Competition and Innovation: While the chip presents an exciting new development, competitors like Nvidia are also continuously innovating. Any potential disruption caused by the Sohu VAPORWARE chip will be met with responses from established players in the AI hardware market.
Energy Efficiency and Scalability: These features are particularly touted as major advantages. However, the practical implications of these benefits will only be clear once the REAL chip is deployed at scale and its performance is observed under diverse operating conditionsβ (MarkTechPost)β.
In summary, while the Sohu chip has the potential to be a significant advancement in AI hardware, its success and the validity of its claims will only be determined over time through extensive real-world testing and adoption.
If you've read this far, read the title and ask yourself, is this CLICKBAIT ?
Google: how many AI startups have failed ?
Do your own research !
Unsubscribed.
If it looks like science fiction itβs maybe because it is. We had a vertical landing lunar module on Texas Instruments some 35 years agoβ¦
I've been thinking for awhile that, because of how popular you are, your tests need randomness, so the tests are fundamentally stable while being situationally chaotic
All of these models are cool, but what can you actually do with them?
How about a tool like LANQL , it actually integrates ChatGPT with your sql databases and lets you talk to your data or directly helps you build your databases.
SQL is Dead!!! Long Live Natural Language!
https://youtu.be/BsUY_Qz9KfA
I mean this is AI news π€―
I'm very surprised these Chinese models are doing so well, and it makes me wonder why.
Was this your first time seeing Falcon Heavy boosters land at the same time?
Etched.
Thinks it csm compete with nvda?
You were excited about Microsofts Recall function??
A good test is to ask an AI model to double a temperature. The correct result involves converting the temperature in degrees (C or F) to kelvin, doubling it, and converting back to degrees. Any AI's I've tested this on just double the initial degrees value, which is incorrect. For instance double 12 degrees C is actually 297.15 degrees C
Jeez, stop the clickbating and the Β«extremeΒ» news announcements
ask the LLM to create 10 annagrams of your firstname lastname
What I want is give me a consumer level $500-$1000 AI Card that is 3-4x as powerful as a 4090 but only for AI tasks. It needs to have like at least 60-80GB of VRAM though. It won't play game etc. It won't even have a HDMI/Display port. I would likely buy one and have a 4080 and one of these in the same PC. I think it's likely possible but I feel like we won't see it though. Not for like 5-10 years.
Need to add a Chinese history test.
Working to ensure "good speak."
How does Cerebras fit in?
Π’Π° ΠΊΠ°ΠΆΠ΅ΡΡΡ Π»ΡΡΡΠ΅ ΡΠ²ΡΠ·ΠΊΠΈ Ρ ΠΈ Π½Π΅ Π²ΠΈΠ΄Π΅Π»)) Π½Ρ 8 ΠΏΡΠΎΡ ΡΡΠΎ ΠΆ Π²Π°ΡΠ΅
Canceled my ChatGPT Plus plan. I am fed up by OpenAI.
Hey bro, I'm Jeremy from an AI company.Would you like to talk about a cooperation?
Etch and qualcomm seem to have better chips than Blackwell, the thing is what kind of time table and volume will they
be able to produce before year end?
8B is not slight bigger than 70B
Maybe the privacy concerns are regarding the data Llama 3 was trained on.
Uh… you're a few years late on those spacex boosters landing at the same time. It's practically a common occurance these days
"Shut up and take my money!!!"
Glad Meta isnβt getting into my iPhone π
Claude sucks, 4o still dominates
If they cancel the deal with Meta over privacy concerns, it means that you were wrong and Elon is right there are still major privacy concerns about the OpenAI deal regardless of how they described it.
One thing I've noticed about the "AI Game" footage is that the observer never turns around or retraces their footsteps at any point.
Are you going to finaly admit R1 Rabbit is a scam or at least acknowledge their HUGE DATA BREACH! or you in their pocket.
You have an awesome channel Matthew!! Really informative and timely. Keep up the good work!!
The idea of managing a programming project without access to the internet and without access to the latest documentation is ludicrous. Until they build that in, it's an interesting toy.
That game looks a lot like an AI video, which is a far cry from an actual game. A game has to have inputs, be performant, responsive, have gameplay, linked sound, keep track of the different player and enemy configurations and states, and the list goes on.
Not that we won't get there, but an actual game that's not just a gimmick made or powered largely by AI is 2 or more decades off. It's not just a matter of capability. It has to be able to run on affordable consumer hardware in numbers that justify the millions of dollars in research and development costs. To say nothing of creating an experience that takes advantage of AI at every level of the game's design, and not just as an NPC simulator
Key points:
– A new AI chip company called Etched claims to have developed a specialized chip called Sohu that can generate over 500,000 tokens per second running the Llama 70B model, outperforming even Nvidia's next-generation Blackwell GPUs.
– OpenAI has announced a delay in the rollout of its advanced voice mode for ChatGPT, which will now be rolled out in alpha to a small group of ChatGPT Plus users in late June, with a full rollout planned for the fall.
– Hugging Face has announced a new open LLM leaderboard that evaluates major open-source language models on various tasks, with the Quen 72B model currently being the top performer.
– Anthropic's Claude 35 Sonet model has secured the number one spot in the coding Arita hard prompts arena and the number two spot in the overall leaderboard by lm.org, surpassing Opus at five times lower cost and being competitive with Frontier models like GPT-4 and Gemini 15 Pro.
– Elon Musk shared an incredible video of two SpaceX rockets landing simultaneously for reuse, showcasing remarkable engineering and scientific achievements.
– A Twitter user shared a video that appears to be a Call of Duty-style game, but is entirely rendered using AI, demonstrating the potential for AI to revolutionize video game development.
– Apple has dropped plans to integrate Meta AI's Llama 3 into Siri over privacy concerns, despite initially exploring integration with multiple AI companies.
Summary made with clipmove.com/chatmove π»
Can you start a ultrasound report question? Or is it too difficult for a non doctor. I use llms to get good ultrasound reports out of one sentence or diagnosis. And I can see huge differences in llm performance and the need for editing
Here is a good one:
say I have 4 points on a coordinate grid: Point a (3|3) point b (6|6), point c (6|3) and point d (2|7). Say I connect point a to point b and point c to point d, and then rotate the shape by 135 degrees by its midpoint. What kind of shape is it, and in what direction would it be?
You could add the "Think step by step to get to the right answer." at the end to make it easier, but still no LLM is able to get this in my testing.
Answer: The resulting shape is a latin cross, like the christian cross. The position of the shape would be would be the normal upright position, so the short line is perfectly horizontal and the long line would be perfectly vertical.
@matthew_berman I have been developing a test; itβs much less trivial than IFEval and all current models fail at it. In fact, it seems OpenAI started detecting and blocking my evals.
When my sub expires for chatGPT, Im done. If there is a company that we shouldnt have at the front of the AI boom, it's them.