Steven Martin

@Teknium @NousResearch I am late to the party but this is yet another banger by Nous Research. This will help me so much as I use different profiles

We are unifying profile management in Hermes Agent starting today. Now the dashboard allows switching management to any of your agent profiles on the machine. No more running multiple dashboards to manage each profile! Also soon, you will only need one gateway to access all of your agents. This will enable us to progress toward full and true isolated multi-tenant solutions in the future.

@intelfabs I think when its not the flavor of the month anymore for CEO's and the boardroom. There will be a reckoning of some sort for AI. Its not going away by a longshot, but someone will need to start rationalizing the spending.

We are very far from AI replacing us Because enterprises can’t quantify AI productivity (only 2% can) If enterprises don’t quantify AI productivity, markets can’t price it. If markets can’t price it, capital stays stuck in infrastructure. If capital stays stuck in infrastructure, the AI rally stays narrow. Nothing changes until enterprises start measuring AI productivity instead of just talking about it

@deforestpeg This is gold!!! Things like this should go viral. Not inane questions asking if you like Fable or 5.5 better than the other!! Great post.

everyone complains about AI api costs. almost nobody optimizes. i kept typing the same 5 fixes in replies so i built the thing that finds them in your actual logs the demo workload (synthetic, 30 days, every inefficiency labeled): $2,330 spend, $1,038 of it recoverable biggest single fix: a 6k-token system prompt billed at full price 24,000 times. one cache_control block serves it at 10% of the price $378 back no llm anywhere in the analysis. every number traces to a formula, and it refuses to extrapolate monthly savings from 3 days of logs because that's marketing, not analysis

@haider1 You can control your costs using any LLM. Here is the magic solution. Don't use it first only use it when you need to.

google and openai are both capable of building a mythos-sized model and achieving mythos-level performance but openai especially chose cost optimization instead, and kept its max model smaller mythos is probably at least 2x the size of gpt-5.5 maybe even 3x

@Jason @Apple It will only be open source until it catches up!

Given token economics, we really need @apple’s new ceo to go all in on workstations that can run local, open source models Ideally, with a router that can flip between local models and frontier models when the former gets stuck. And America needs an open source champion — we really should not be comfortable with the Chinese owning the open source LLM market to the extent they do

@Beth_Kindig Nvidia would be crazy to not hedge and have a backup plan. Forget the talk about China, Taiwan is in the ring of fire, large earthquakes have consistently happened here for a millennium. They should be hedging with Intel and Samsung, maybe even Rapidus.

Nvidia is reportedly testing Intel’s 18A for manufacturing a new design integrating four GPU dies into a single package. $NVDA $INTC $TSM $AMD

$BIIB Money maker?? Let me tell you. BIIB is the most purchased stock in the last 30 days by members of Congress. Guess what happens on August 24. Biogen has an "imminent FDA decision that day" I wonder what they know that we don't :P

@_JKNFT_ @TTrimoreau Sad thing is they give you more job security. Who else is going to fix that MVP release the boss loves?

@TTrimoreau "People who are building stuff in Claude code without proper coding knowledge are a threat to me." There, I fixed it for you.

People who are building stuff in Claude code without proper coding knowledge scares me.

@IanCutress @Index_shu Yes, indeed how do you miss Rapidus?

@Index_shu What about Rapidus ?!?

これが日本を代表する新たな半導体企業群です

@SemiAnalysis_ If there was every a no shit sherlock moment this would be it. SemiAnalysis: Subscriptions save you lots of money. Me: you didn't think i know that?

Recently, we purchased one of each Anthropic/OpenAI subscription plan and randomly ran long horizon coding tasks until we exhausted the weekly limit. It's widely believed that a $200/month plan maxes out at ~$2000/month worth of tokens (assuming API pricing). However, we found that the subscriptions are actually far more generous. (2/4)

What's the better business model for an AI lab, subscription or API? (1/4)🧵

$25.17 cents vs $833.34 I know what model i am trying first. Guess what if its not good enough you can still spend your money on Fable or Chat GPT maybe less if you input what DeepSeek came up with.

I see a change bumbling through the cracks. I see this talk about Fable capability, but who can afford it. I see the calls for Open Ai to release GPT 5.6 even if its "a little bit worse than Fable 5. Newsflash GPT 5.5 is considerably more expensive. So you are now "hoping" the most expensive model of all drop their price? Guess what a 300% price drop on output tokens puts it right around the cost of Fable 5 right now. Please come join the common users with Kimi 2.6 and DeepSeek V4. Make sure you take 10 minutes to tailor your prompts to for each model. I think you will be surprised. I believe that most people could reduce 75% of their usage or more to one of those 2 models. Tell me what you think or what your doing to control costs.

@Pablo01618 They just might, if Oracle asks! You don't think Larry Ellison has a few politicians on speed dial? I do.

What if the fed came in and backstopped $ORCL in our Wall Thesis $ORCL plays a pivotal roll in the buildout and eats at every single substrate layer, as far as bang for your buck thats where I would be punting if I was the federal government

@WayWLeung Great post Chuip! You nailed it slowly but surely Lip-Bu Tan is winning them over. He is doing it the exactly the way he did at Cadence. He is a steady leader that practices what he preaches. Wall Street has developed a trust with him over the years

$intc 美银对Intel的最新评级大幅改变（从敌视改成买入），当三个最顽固最敌视Intel的Wall Street巨头投行改变了态度时，Intel重新估值的趋势已经势不可挡，Intel foundry会重新带Intel站立回去半导体王者地位，成为美国和世界AI行业核心的一环。 Boa的态度改变再一次说明了在AI年代，真的没有必要再去迷信分析师，自己的研究更加重要，这些分析师偏见和敌视以及滞后都会让投资者失去先机。 恭喜各位早早投资Intel的朋友，拿着低成本（我均价$18.64）一直持有，伴随着Intel的重新崛起而不断获得丰厚回报

@Bhavani_00007 None of the above I use OpenCode Go and ChatGpt plus plan. That covers most of my needs. If I need more I directly use DeepSeek api

developers, which one are you choosing? $200 Claude Code or $100 Codex + $100 Claude Code

@imnotharsh 30 year partnership renewed with fresh ideas for the Ai boom. Another win for Intel.

JUST IN: $INTC Intel and $HPE Hewlett Packard Enterprise Company expand 30 year partnership to deliver Enterprise IT infrastructure across on‑premises, hybrid, and edge environments At HPE Discover 2026, where Intel serves as a Visionary sponsor, the long-standing alliance will emphasize integrated infrastructure solutions designed for enterprise modernization, practical artificial intelligence deployment, and secure hybrid operations from edge to cloud. The primary focus will center on HPE ProLiant Compute Gen12 servers powered by Intel Xeon 6 processors. These platforms deliver measurable advancements in performance, including up to 2.5 times higher high-performance computing throughput in relevant workloads, alongside improved efficiency measured in performance per watt and per rack. The processors support CPU-based AI inferencing for practical enterprise use cases, such as small language models and high-performance LLM inference through validated vLLM integrations, without requiring accelerator hardware in every deployment. Security enhancements will receive prominent attention. The solutions incorporate Intel technologies that provide quantum-resistant cryptography aligned with NIST and CNSA 2.0 standards, Intel Software Guard Extensions for confidential computing, and Trust Domain Extensions for virtual machine isolation. These features address compliance and data protection requirements in regulated and mission-critical environments. Edge and telecommunications workloads will feature through HPE Edgeline systems combined with Intel Xeon 6 capabilities. The partnership will highlight reduced power consumption, smaller deployment footprints, and improved network performance for 5G core and distributed AI scenarios. This supports organizations managing growing data volumes outside traditional data centers. Hybrid cloud and virtualization offerings will also be showcased. These include HPE GreenLake for flexible consumption-based models, validated Azure Local solutions, VMware trusted platforms, and modern VM management through HPE Morpheus VM Essentials on Intel architecture. Broader ecosystem elements, such as HPE Alletra storage and Juniper networking integrations, will demonstrate end-to-end optimized stacks for AI, analytics, and industry-specific applications. Customer benefits will be framed around total cost of ownership reductions, accelerated time to value, sustainability improvements, and simplified compliance. Case studies and demonstrations in HPE Customer Innovation Centers, along with on-demand resources, will illustrate real-world outcomes in data center refresh, edge digitization, and AI readiness. The messaging positions the partnership as a foundation for organizations seeking balanced, silicon-rooted performance and security while advancing practical AI adoption across core and distributed environments. Attendees can expect technical sessions, solution briefs, and interactive exhibits reinforcing these capabilities.

@usr_bin_roygbiv Not such a bargain then. So Much for that idea

@Colosteve2000 not right now under any circumstances the cost to rent for 1 year is 80% of the cost to buy new

Hot take: They're not subsidized their margins are insane. They are just absolutely raping api customers. Anyone who has used deepseek or hosted anything and done the math on hardware/power costs knows this

@demishassabis @bodonoghue85 Now imagine this 4x faster using Cerebra’s technology and even more crazy add @TileRT_AI. Obviously you aren’t likely to need both but imagine the speed!

Awesome to see this innovation in text diffusion. DiffusionGemma is lightning fast, 4x faster than other Gemma 4 models! Congrats to @bodonoghue85 and the team who worked so hard on this - excited to see what people build with it!