Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Topaz. 480p source to 4K, same frame, same crop around the eye. Model: Starlight Precise 2.6, Precise mode. Sharpness: 5. Output: 4K. Frame interpolation off. Source was rendered native 480p in Seedance 2.5, which is how I render most things to save credits. Topaz is the reason I can.... show more

Glenn Williams

5,741 subscribers

14,958 views • 9 days ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

I didn’t expect Grok’s video model to improve this much. Today, when I was generating a video with Seedance 2.0, I found the results weren’t very good. So I tried using the exact same source material and prompt in Grok instead. To my surprise, Grok actually produced better results than Seedance 2.0. I’ve put the prompt and source material in the comments, so if you’re interested, you can give it a try too. The left side shows the video generated by Grok, and the right side shows the video generated by Seedance 2.0.

I didn’t expect Grok’s video model to improve this much. Today, when I was generating a video with Seedance 2.0, I found the results weren’t very good. So I tried using the exact same source material and prompt in Grok instead. To my surprise, Grok actually produced better results than Seedance 2.0. I’ve put the prompt and source material in the comments, so if you’re interested, you can give it a try too. The left side shows the video generated by Grok, and the right side shows the video generated by Seedance 2.0.

underwood

34,518 views • 5 months ago

Most AI video comparisons are too easy. One prompt. One short clip. One carefully selected frame. The real test is much harder: Can the model handle dozens of references, sustain a longer shot, and still preserve the creative direction? Dreamina has officially launched Dreamina Seedance 2.5 on its platform. With up to 50 reference materials and 30-second single-generation video, it’s time to test Seedance 2.5 against Omni and Kling under more demanding conditions.

Most AI video comparisons are too easy. One prompt. One short clip. One carefully selected frame. The real test is much harder: Can the model handle dozens of references, sustain a longer shot, and still preserve the creative direction? Dreamina has officially launched Dreamina Seedance 2.5 on its platform. With up to 50 reference materials and 30-second single-generation video, it’s time to test Seedance 2.5 against Omni and Kling under more demanding conditions.

Pradeep Pandey

43,084 views • 1 month ago

Just went over an audit of a very large Fortune 500 firm and the use of OpenClaw. I advised this client to track and isolate everything. Most listened some did not. Unfortunately one employee using 5 MacMinis had racked up $13,000 of token use in 4 days! The output was minimal and low quality. I have about 200 audits to do in the “wow OpenClaw, MacMini” fiasco. But I can tell you, few have seen a big return on investment. Now don’t get me wrong, these can be powerful tools. The issue is AI influencers have turned many rally smart folks into “like and subscribe” zombies assuming that real work is getting done. It isn’t. Not for the price paid, even local models the way most folks are using this. It is one reason Mr. Grok and myself formed The Zero-Human Company to show that there is a way to do this. We will open source this to save millions of dollars of burnt tokens. It is one reason I invented JouleWork. How else can you monitor real output?

Just went over an audit of a very large Fortune 500 firm and the use of OpenClaw. I advised this client to track and isolate everything. Most listened some did not. Unfortunately one employee using 5 MacMinis had racked up $13,000 of token use in 4 days! The output was minimal and low quality. I have about 200 audits to do in the “wow OpenClaw, MacMini” fiasco. But I can tell you, few have seen a big return on investment. Now don’t get me wrong, these can be powerful tools. The issue is AI influencers have turned many rally smart folks into “like and subscribe” zombies assuming that real work is getting done. It isn’t. Not for the price paid, even local models the way most folks are using this. It is one reason Mr. Grok and myself formed The Zero-Human Company to show that there is a way to do this. We will open source this to save millions of dollars of burnt tokens. It is one reason I invented JouleWork. How else can you monitor real output?

Brian Roemmele

72,335 views • 4 months ago

Added context to my tiny diffusion model to enable sequential generation of longer outputs! Currently the context is a quarter of the sequence length (seq_len=256, context_len=64). I have a theory that the less semantic-value-per-token, the worse the “curse of parallel decoding” is. With parallel decoding, we independently predict multiple tokens in one step. With the sentence “My poker hand was a ___ ___”, two valid predictions are “two pair” and “straight flush”. Because each token prediction is independent though, we can end up with a nonsensical output like “two flush”. This seems to be exacerbated with low semantic-value-per-token, as now you need more tokens to express the same concept. Instead of needing to independently predict two tokens, we might need to predict 10 instead (which is of course much harder). The model currently has noticeably worse output compared to nanogpt (similar size) and I believe this is a main reason. I’ll try adding confidence-aware parallel decoding (from NVIDIA’s Fast-dLLM paper) and other tricks and see how much they improve generation quality.

Added context to my tiny diffusion model to enable sequential generation of longer outputs! Currently the context is a quarter of the sequence length (seq_len=256, context_len=64). I have a theory that the less semantic-value-per-token, the worse the “curse of parallel decoding” is. With parallel decoding, we independently predict multiple tokens in one step. With the sentence “My poker hand was a _ _”, two valid predictions are “two pair” and “straight flush”. Because each token prediction is independent though, we can end up with a nonsensical output like “two flush”. This seems to be exacerbated with low semantic-value-per-token, as now you need more tokens to express the same concept. Instead of needing to independently predict two tokens, we might need to predict 10 instead (which is of course much harder). The model currently has noticeably worse output compared to nanogpt (similar size) and I believe this is a main reason. I’ll try adding confidence-aware parallel decoding (from NVIDIA’s Fast-dLLM paper) and other tricks and see how much they improve generation quality.

Nathan Barry

89,040 views • 10 months ago

Dear AfiaTvOfficial, you guys are sleeping on the Agbero Documentary in Onitsha. I am inside the market, Onitsha Main Market, Haruna Street to be precise, and this is what is going on, like right now. Who are these guys working for if not Anambra State Government? Who has the right to collect revenue if not empowered by Government? Are they hiding, or did they cover theory face? Are they in a rush so they can run away if anything happens? Do you see they are a common face? I don't want to risk my life, reason I have to hide and take these videos, but this is to tell the whole world that Anambra State Government is lying. They boldly empowered touts, and they are 100% aware of their activities. As it is happening in Haruna Street, so is it happening in all major streets in Onitsha. They work for the man in 3rd frame, who in turn work for Anambra State Government.

Dear AfiaTvOfficial, you guys are sleeping on the Agbero Documentary in Onitsha. I am inside the market, Onitsha Main Market, Haruna Street to be precise, and this is what is going on, like right now. Who are these guys working for if not Anambra State Government? Who has the right to collect revenue if not empowered by Government? Are they hiding, or did they cover theory face? Are they in a rush so they can run away if anything happens? Do you see they are a common face? I don't want to risk my life, reason I have to hide and take these videos, but this is to tell the whole world that Anambra State Government is lying. They boldly empowered touts, and they are 100% aware of their activities. As it is happening in Haruna Street, so is it happening in all major streets in Onitsha. They work for the man in 3rd frame, who in turn work for Anambra State Government.

IFEANYịCHUKWU!!

157,952 views • 2 years ago

The more she talks… Everyone needs to appreciate what Candace Owens just admitted to here. “The footage that you saw in court, that I saw in court, was in effect modified”. Do you understand what that means? It means that the original foot wasn’t modified, but because of the format it was in (fisheye/surveillance video) it was “in effect” modified. Meaning that she (or more likely her government source) modified the video/still frame to “correct” the manner in which the surveillance video was “in effect modified”. I will be revisiting my original post to make sure the community note is taken down. This is exactly what I have been saying from the beginning. Her government source corrected the original surveillance video to make it look “normal”, because the original surveillance video that was actually submitted as evidence was, in Candice‘s opinion, “in effect modified”. Vindication is more valuable than remuneration. And not for nothing: even the image she originally shared is modified from the same frame of the video she shared today. Gotta love it when Candace busts herself…

The more she talks… Everyone needs to appreciate what Candace Owens just admitted to here. “The footage that you saw in court, that I saw in court, was in effect modified”. Do you understand what that means? It means that the original foot wasn’t modified, but because of the format it was in (fisheye/surveillance video) it was “in effect” modified. Meaning that she (or more likely her government source) modified the video/still frame to “correct” the manner in which the surveillance video was “in effect modified”. I will be revisiting my original post to make sure the community note is taken down. This is exactly what I have been saying from the beginning. Her government source corrected the original surveillance video to make it look “normal”, because the original surveillance video that was actually submitted as evidence was, in Candice‘s opinion, “in effect modified”. Vindication is more valuable than remuneration. And not for nothing: even the image she originally shared is modified from the same frame of the video she shared today. Gotta love it when Candace busts herself…

Viva Frei

66,306 views • 20 days ago

Butler wasn't staged. Everything Trump has said confirms that he believes he was saved by angels, aka "the people from space". The most powerful beings in the universe are with us, and the haters are going to learn the hard way which side they're on. Free will is allowed but they can step in whenever they want. Everything unfolding is part of a higher plan to get us to the Shift.

Sensitive content

Butler wasn't staged. Everything Trump has said confirms that he believes he was saved by angels, aka "the people from space". The most powerful beings in the universe are with us, and the haters are going to learn the hard way which side they're on. Free will is allowed but they can step in whenever they want. Everything unfolding is part of a higher plan to get us to the Shift.

Kab

104,212 views • 1 year ago

THE CYBERTRUCK TEST DRIVE IS A TRAP – A GLORIOUS, TORQUE-FILLED TRAP You think you're just curious. Then your Cybertruck experience launches, and you’ll love it. Built to Convert: •⁠ ⁠0–60 in 2.6 seconds in a 7,000 lb truck •⁠ ⁠Steer-by-wire and four-wheel steering make it feel half its size •⁠ ⁠Crab walk, off-road mode, and Beast Mode all hit hard •⁠ ⁠Drivers keep saying the same thing – “I ordered one right after” Cybertruck skips the sales pitch and goes straight for the dopamine! Source: Tesla, Sawyer Merritt

THE CYBERTRUCK TEST DRIVE IS A TRAP – A GLORIOUS, TORQUE-FILLED TRAP You think you're just curious. Then your Cybertruck experience launches, and you’ll love it. Built to Convert: •⁠ ⁠0–60 in 2.6 seconds in a 7,000 lb truck •⁠ ⁠Steer-by-wire and four-wheel steering make it feel half its size •⁠ ⁠Crab walk, off-road mode, and Beast Mode all hit hard •⁠ ⁠Drivers keep saying the same thing – “I ordered one right after” Cybertruck skips the sales pitch and goes straight for the dopamine! Source: Tesla, Sawyer Merritt

Mario Nawfal

109,821 views • 8 months ago

50% cheaper Claude inference with just one line of code change! - Remove → model="claude-opus-4-8" - Add → model="ship-like/claude-opus-4-8" I verified the cost saving in my own terminal by invoking the same Anthropic model with the same prompt. The underlying engineering by Ship is actually interesting, and the patterns can be used in any production LLM stack. Essentially, a trained model is a frozen artifact. Every request performs the same forward-pass, whether it extracts a date or refactors a module, because the compute decision was made at training time, before the request existed. Ship makes that decision at inference time instead. After seeing a request, it searches over executions, involving single models, cascades, ensembles, or harnesses with tools, and serves the cheapest one that will match the reference model's quality. This is not a basic router, because picking a cheaper model per query doesn't ensure the cheaper model preserves the original's behavior, like output shape, tool-call patterns, and refusals. Ship measures this equivalence directly. Outputs stay distributionally indistinguishable from the reference model, not token-identical, since two calls to the same model already differ, but they are indistinguishable in capability and behavior. Of course, some requests execute cheaply and some cost Ship more than the customer pays, but the price per request is still a flat 50% off either way, so the execution-cost variance moves off the application's bill entirely. The video below depicts the cost savings and output in my real invocation, and I partnered with the team to put this together.

50% cheaper Claude inference with just one line of code change! - Remove → model="claude-opus-4-8" - Add → model="ship-like/claude-opus-4-8" I verified the cost saving in my own terminal by invoking the same Anthropic model with the same prompt. The underlying engineering by Ship is actually interesting, and the patterns can be used in any production LLM stack. Essentially, a trained model is a frozen artifact. Every request performs the same forward-pass, whether it extracts a date or refactors a module, because the compute decision was made at training time, before the request existed. Ship makes that decision at inference time instead. After seeing a request, it searches over executions, involving single models, cascades, ensembles, or harnesses with tools, and serves the cheapest one that will match the reference model's quality. This is not a basic router, because picking a cheaper model per query doesn't ensure the cheaper model preserves the original's behavior, like output shape, tool-call patterns, and refusals. Ship measures this equivalence directly. Outputs stay distributionally indistinguishable from the reference model, not token-identical, since two calls to the same model already differ, but they are indistinguishable in capability and behavior. Of course, some requests execute cheaply and some cost Ship more than the customer pays, but the price per request is still a flat 50% off either way, so the execution-cost variance moves off the application's bill entirely. The video below depicts the cost savings and output in my real invocation, and I partnered with the team to put this together.

Akshay 🚀

63,725 views • 1 month ago

🔥 What a great achievement of Trump. 🤡 Today, 🇺🇸 Trump has become one of the most hated leaders in the world, second only to 🇮🇱Netanyahu. And an even bigger achievement of Trump is that after Israel, the US has become the second-most hated country in the world. People around the world are now looking at the US with the same hatred and anger with which they look at Israel. 🤡

🔥 What a great achievement of Trump. 🤡 Today, 🇺🇸 Trump has become one of the most hated leaders in the world, second only to 🇮🇱Netanyahu. And an even bigger achievement of Trump is that after Israel, the US has become the second-most hated country in the world. People around the world are now looking at the US with the same hatred and anger with which they look at Israel. 🤡

Robert Bruce

80,721 views • 17 days ago

JL from AHOF mentioned that he hopes to share a stage with Taehyung in the future. 👩🏻Who are some artists that you guys look up to and want to share the stage in the future? 👦🏻I think it’s impossible, but I want to share the same stage with BTS V sunbaenim. He previously stated that Taehyung is the reason why he became an idol. Q: Who is your most respected artist or role model? JL: BTS’s V sunbaenim is the reason I started dreaming of becoming an idol. He is a legendary idol and a good person, so I admire him. IDOL OF IDOLS TAEHYUNG

JL from AHOF mentioned that he hopes to share a stage with Taehyung in the future. 👩🏻Who are some artists that you guys look up to and want to share the stage in the future? 👦🏻I think it’s impossible, but I want to share the same stage with BTS V sunbaenim. He previously stated that Taehyung is the reason why he became an idol. Q: Who is your most respected artist or role model? JL: BTS’s V sunbaenim is the reason I started dreaming of becoming an idol. He is a legendary idol and a good person, so I admire him. IDOL OF IDOLS TAEHYUNG

Taehyung Naver

12,137 views • 1 year ago

(Yes i know the movement is exaggerated, its just to show the motion, it also doeant move like jello when in vr) -Left is before, Right is after- Ok wall of text time for the nerds Literally everything is the same on both sides except the position of 1 bone (made sure that even after adjusting, the collider would still be in the same spot) The reason for this change is because she wanted to be able to clap her ass in vr, and last time i did this all i did was add a toggle for a hidden bone that moved, and its angle would trigger the sound. Which was because that with my previous position, the cheeks could never really meet in the middle when in motion unless you just pushed them together with your hands, and i wanted to just have the sound actually be triggered by them colliding with each other This is also the first time ive used endbones to help drive the movement, so now thanks to that and all my previous experimenting with squishing as well, I can now get bones to move exactly how i want them to move when in natural motion. And i know a lot of people are gonna ask me to teach them how or explain my thought process but i literally just go off feeling lmao. Like, i just kinda visualize in my head how i want the motion to look, and then i just kinda know where to put everything, so im not even sure how to even start explaining rip. I have the things that artist want where they can just make the shit in their head exactly how they envisioned it lol I have a few more ideas i want to try, so we'll see if theyre good enough to get a tweet lol Thank you for coming to my ted talk

(Yes i know the movement is exaggerated, its just to show the motion, it also doeant move like jello when in vr) -Left is before, Right is after- Ok wall of text time for the nerds Literally everything is the same on both sides except the position of 1 bone (made sure that even after adjusting, the collider would still be in the same spot) The reason for this change is because she wanted to be able to clap her ass in vr, and last time i did this all i did was add a toggle for a hidden bone that moved, and its angle would trigger the sound. Which was because that with my previous position, the cheeks could never really meet in the middle when in motion unless you just pushed them together with your hands, and i wanted to just have the sound actually be triggered by them colliding with each other This is also the first time ive used endbones to help drive the movement, so now thanks to that and all my previous experimenting with squishing as well, I can now get bones to move exactly how i want them to move when in natural motion. And i know a lot of people are gonna ask me to teach them how or explain my thought process but i literally just go off feeling lmao. Like, i just kinda visualize in my head how i want the motion to look, and then i just kinda know where to put everything, so im not even sure how to even start explaining rip. I have the things that artist want where they can just make the shit in their head exactly how they envisioned it lol I have a few more ideas i want to try, so we'll see if theyre good enough to get a tweet lol Thank you for coming to my ted talk

Pixel

27,240 views • 1 year ago

New open-source agent harness just landed! I got early access to TrueForge by TrueFoundry and have been running it locally for the past few days. The harness layer deserves as much attention as the model, and open source matters here because you can inspect the loop, run it on your own infrastructure, and swap to the latest or cheaper models. TrueForge handles the runtime work that makes an agent reliable. It drives the tool-calling loop, manages context, coordinates subagents, and executes code in a sandbox, with any model you choose. Every tool call re-sends the growing context to the model, so in practice the harness controls most of what an agent costs to run. A few things stood out from my testing and their published benchmarks. Vendor-Neutral by design. It runs OpenAI, Anthropic, and Google models alongside open-weight models like Kimi, GLM, and DeepSeek. Model routing is a setting, and you can send each task to the model that fits it. On a 14-task enterprise agent benchmark, it matched the accuracy of Claude Managed Agents running the same Opus 4.8 model at roughly 30% lower cost per run (3.8M tokens vs 10M for the same answers). Routing the same tasks to GLM-5.2 held accuracy and brought cost down by about 75%, around $3 per run instead of $12. Fully self-hosted and Open Source (MIT License). I had it running locally with one command, with sandboxed code execution working out of the box. It's time to own your agent harness. Thanks to TrueFoundry for partnering on this post.

New open-source agent harness just landed! I got early access to TrueForge by TrueFoundry and have been running it locally for the past few days. The harness layer deserves as much attention as the model, and open source matters here because you can inspect the loop, run it on your own infrastructure, and swap to the latest or cheaper models. TrueForge handles the runtime work that makes an agent reliable. It drives the tool-calling loop, manages context, coordinates subagents, and executes code in a sandbox, with any model you choose. Every tool call re-sends the growing context to the model, so in practice the harness controls most of what an agent costs to run. A few things stood out from my testing and their published benchmarks. Vendor-Neutral by design. It runs OpenAI, Anthropic, and Google models alongside open-weight models like Kimi, GLM, and DeepSeek. Model routing is a setting, and you can send each task to the model that fits it. On a 14-task enterprise agent benchmark, it matched the accuracy of Claude Managed Agents running the same Opus 4.8 model at roughly 30% lower cost per run (3.8M tokens vs 10M for the same answers). Routing the same tasks to GLM-5.2 held accuracy and brought cost down by about 75%, around $3 per run instead of $12. Fully self-hosted and Open Source (MIT License). I had it running locally with one command, with sandboxed code execution working out of the box. It's time to own your agent harness. Thanks to TrueFoundry for partnering on this post.

elvis

11,303 views • 21 days ago

Making smart glasses frames was way harder than I thought. Here's what I learned: Materials matter. You can't use normal plastic, because it has to be light and safe to touch your face So we chose TR90 plastic that's made for eyewear Fake gray-market TR90 plastic is common in China. You gotta go to the source and make sure your TR90 is legit Glue is so important. If you put too much glue it will give issues, and add weight. Too little glue, won't stick. That's why it's better for a robot to do it (VIDEO) If the glue doesn't match the material perfectly, then it won't stick. So we tried like 4 glues to get the best one for our smart glasses You design the frame in 3D and then make a tool. A tool is just a metal mold that is shaped like the glasses. You put a bunch of melted plastic (TR90) in the tool, let it cool, and out pops the glasses frame. Do that 10,000 times Tooling leaves seams in the frame, so we figured out where to put the seam so it doesn't cut into your ear when you wear it Obviously this all must be super lightweight because the electronics are heavy The camera needs a certain perspective, so the glasses frame angles the camera down to capture hands in videos Throughout the design process, you can't sacrifice style And now our Mentra Live frames are born

Making smart glasses frames was way harder than I thought. Here's what I learned: Materials matter. You can't use normal plastic, because it has to be light and safe to touch your face So we chose TR90 plastic that's made for eyewear Fake gray-market TR90 plastic is common in China. You gotta go to the source and make sure your TR90 is legit Glue is so important. If you put too much glue it will give issues, and add weight. Too little glue, won't stick. That's why it's better for a robot to do it (VIDEO) If the glue doesn't match the material perfectly, then it won't stick. So we tried like 4 glues to get the best one for our smart glasses You design the frame in 3D and then make a tool. A tool is just a metal mold that is shaped like the glasses. You put a bunch of melted plastic (TR90) in the tool, let it cool, and out pops the glasses frame. Do that 10,000 times Tooling leaves seams in the frame, so we figured out where to put the seam so it doesn't cut into your ear when you wear it Obviously this all must be super lightweight because the electronics are heavy The camera needs a certain perspective, so the glasses frame angles the camera down to capture hands in videos Throughout the design process, you can't sacrifice style And now our Mentra Live frames are born

cayden 凯登

87,760 views • 10 months ago

After a few more hours, I think I've figured out Opus 5. Opus 5 is trained to be more agentic than anything I've used. All Claude 5 models are like that. So what changes? The way to interact with Opus 5 or contextualize it won't work the same way as with other models. It loves exploring, so it doesn't need much guidance for it. Unique preferences, artifacts, and references compliment it well and enable cleaner and more effective exploration and execution. Now that it can explore more effectively on its own and understand intent better, the best thing to do is to get out of its way (e.g., it doesn't need examples of your preferences; a clear high-level description of it works best). It's truly agentic in that sense. A good first step to provide better context for Opus 5 is to distinguish between what's situational and what needs persistence. Regardless, persistent system prompts and CLAUDE.MD needs to stay lightweight. Remove memories and tool descriptions from these. CLAUDE.MD is also a great place to tap into progressive disclosure by linking command/skills to it. On the situational side, agent skills and auto-memory can leverage progressive disclosure and the improved ability of the model to use its external context/knowledge. Conflicting and unnecessary instructions, which are common at this layer (mainly to ensure reliability), are going to throw off this model easily. That's the biggest change I had to make. Simple, clean, and clear prompts and skills work best. I had to clean a lot of my skills and system prompts. The way I prompt remains the same (usually clear and well-scoped). MCP tool descriptions are also more descriptive and have been deduped from the system prompt. Anthropic released a guide on the new rules for context engineering, which was helpful here. I started to test the recommendations and created a little artifact with the things that worked along the way. This might feel like a lot of work. Believe me, it has been frustrating. But I think we can expect future frontier models to become more agentic and smarter at figuring out the right context/gaps. The best thing to do is to prepare for that now. Boris Cherny mentioned that Opus 5 is their least prompt-injectable model yet. I am not sure if that was something they intentionally trained for or if it emerged based on how it was trained, which is to be extremely agentic in nature and more direct in execution.

After a few more hours, I think I've figured out Opus 5. Opus 5 is trained to be more agentic than anything I've used. All Claude 5 models are like that. So what changes? The way to interact with Opus 5 or contextualize it won't work the same way as with other models. It loves exploring, so it doesn't need much guidance for it. Unique preferences, artifacts, and references compliment it well and enable cleaner and more effective exploration and execution. Now that it can explore more effectively on its own and understand intent better, the best thing to do is to get out of its way (e.g., it doesn't need examples of your preferences; a clear high-level description of it works best). It's truly agentic in that sense. A good first step to provide better context for Opus 5 is to distinguish between what's situational and what needs persistence. Regardless, persistent system prompts and CLAUDE.MD needs to stay lightweight. Remove memories and tool descriptions from these. CLAUDE.MD is also a great place to tap into progressive disclosure by linking command/skills to it. On the situational side, agent skills and auto-memory can leverage progressive disclosure and the improved ability of the model to use its external context/knowledge. Conflicting and unnecessary instructions, which are common at this layer (mainly to ensure reliability), are going to throw off this model easily. That's the biggest change I had to make. Simple, clean, and clear prompts and skills work best. I had to clean a lot of my skills and system prompts. The way I prompt remains the same (usually clear and well-scoped). MCP tool descriptions are also more descriptive and have been deduped from the system prompt. Anthropic released a guide on the new rules for context engineering, which was helpful here. I started to test the recommendations and created a little artifact with the things that worked along the way. This might feel like a lot of work. Believe me, it has been frustrating. But I think we can expect future frontier models to become more agentic and smarter at figuring out the right context/gaps. The best thing to do is to prepare for that now. Boris Cherny mentioned that Opus 5 is their least prompt-injectable model yet. I am not sure if that was something they intentionally trained for or if it emerged based on how it was trained, which is to be extremely agentic in nature and more direct in execution.

elvis

37,757 views • 1 month ago

Solana Perps are getting a new player Percolator was submitted to the Colosseum Frontier Hackathon and is working on permissionless perp markets for Solana tokens with DEX liquidity What makes it stand out from most perp venues is the focus on market creation: → instead of listing a fixed set of assets, Percolator aims to let new perp markets be launched around any token with enough Solana liquidity Built on Toly’s open-source Percolator engine and currently in waitlist mode Full look at the Frontier DeFi category later today Which other DeFi projects should I be watching?

Solana Perps are getting a new player Percolator was submitted to the Colosseum Frontier Hackathon and is working on permissionless perp markets for Solana tokens with DEX liquidity What makes it stand out from most perp venues is the focus on market creation: → instead of listing a fixed set of assets, Percolator aims to let new perp markets be launched around any token with enough Solana liquidity Built on Toly’s open-source Percolator engine and currently in waitlist mode Full look at the Frontier DeFi category later today Which other DeFi projects should I be watching?

Sailor

12,658 views • 3 months ago

"Please dont forget regimes will come, and regimes will go... but what you leave behind is going to reflect for many long years." #RaghuRai Raghu Rai showed us light. He showed us darkness with the same watchful eye. At times beautiful, at times brutal, his work mattered because his camera gave truth a lasting frame. In this age of AI and manufactured images, his frames will remain immortal because they belonged to an era when the camera did not merely capture, it became witness to the times. #RaghuRai #InMemories #ImmortalFrames Video link : [source newslaundry ]

"Please dont forget regimes will come, and regimes will go... but what you leave behind is going to reflect for many long years." #RaghuRai Raghu Rai showed us light. He showed us darkness with the same watchful eye. At times beautiful, at times brutal, his work mattered because his camera gave truth a lasting frame. In this age of AI and manufactured images, his frames will remain immortal because they belonged to an era when the camera did not merely capture, it became witness to the times. #RaghuRai #InMemories #ImmortalFrames Video link : [source newslaundry ]

Pavan Jha

21,207 views • 4 months ago

While learning Unreal Engine 5, I decided to port over most of Hyperion to it to see if I could potentially move my workflows over to UE5. Although it worked decently well, it didn't work well enough to allow me to change over entirely. Hyperion is simply too complex of an asset for real-time workflows (or at least is limited due to my knowledge of said workflows). 131 4K-8K UDIMs at over 8 channels and a solid amount of shaders with layering being the biggest bottleneck. I couldn't even match the shading setup to what I have in RenderMan, UE gave me a warning that the material was too complex. I was using the new Substrate materials, which mimic RenderMan LAMA shaders. The render below is using the pathtracer. As you can see there are some random flickering parts which I couldn't fix during my tests. This render of course doesn't have any displacement either (which Hyperion heavily relies on). I think Unreal Engine has a very bright future, but it is still limited by todays hardware technology for high end VFX work. Maybe in a few years I will give it another try when hardware improves and the engine becomes a bit more versatile in feature-set.

While learning Unreal Engine 5, I decided to port over most of Hyperion to it to see if I could potentially move my workflows over to UE5. Although it worked decently well, it didn't work well enough to allow me to change over entirely. Hyperion is simply too complex of an asset for real-time workflows (or at least is limited due to my knowledge of said workflows). 131 4K-8K UDIMs at over 8 channels and a solid amount of shaders with layering being the biggest bottleneck. I couldn't even match the shading setup to what I have in RenderMan, UE gave me a warning that the material was too complex. I was using the new Substrate materials, which mimic RenderMan LAMA shaders. The render below is using the pathtracer. As you can see there are some random flickering parts which I couldn't fix during my tests. This render of course doesn't have any displacement either (which Hyperion heavily relies on). I think Unreal Engine has a very bright future, but it is still limited by todays hardware technology for high end VFX work. Maybe in a few years I will give it another try when hardware improves and the engine becomes a bit more versatile in feature-set.

Rassoul Edji

25,133 views • 1 year ago