researchCobra

Armchair Analyst - L/S Equities & AI
"historia magistra vitae"

[Mia @MiaAI_lab]

Best local AI models for 8-256gb of VRAM

8gb VRAM: Qwen3.8-27B IQ1_S
https://t.co/ra9r2zUJTy

16gb VRAM: Qwen3.8-27B Q2_K_XL
https://t.co/07HdkXRimI

RTX 3090: Qwen3.8-27B Q4_K_M
https://t.co/M50sN4X40z

RTX 5090: Qwen3.8-27B NVFP4
https://t.co/y2vY62rE1I

RTX PRO 6000: Qwen3.8-27B NVFP4
https://t.co/E1e2PuWa8l

DGX Spark: DeepSeek v4 Flash / Qwen3.8-27B
https://t.co/531EaR6FWJ
https://t.co/rooS9ykL91

2x DGX Sparks: DeepSeek v4 Flash
https://t.co/6HXa9pxqhj

Qwen3.8-27B and DeepSeek v4 Flash are running the show, but this might change soon 👀

[researchCobra @research_cobra]

If you are in fundamnetal L/S and open to experiement with AI now is the time to try Hermes for free with almost frontier intelligence

Find Hermes impressive for second-brain and research-fan outs that dig independently. Hermes also has massive toolkit to deploy for such tasks

[Teknium 🪽 @Teknium]

Maybe its 10 quadrillion the world may never know!

[researchCobra @research_cobra]

@mr_r0b0t Do you prefer Ling to Deepseek v4 / latest Qwen? If so why?

[researchCobra @research_cobra]

@u1tra_instinct @mr_r0b0t LFG!

[researchCobra @research_cobra]

@kimmonismus @davis7 Anthropic/OpenAI IPO will go up in flames if true

[researchCobra @research_cobra]

@MiaAI_lab Working on recreating your results. When working with it do you run it with or without thinking for actual work tasks?

[researchCobra @research_cobra]

This is a gamechanger for the single DGX community - congrats @MiaAI_lab running tests now

[Mia @MiaAI_lab]

A BIG moment for all DGX Spark users ⚡️

You can now run DeepSeek v4 Flash 0731 without needing a second unit, with quality high enough for reliable code generation, high context, and great speed!

Optimized for single stream session:
- EXL3 quantization
- 384k context (conservative) / ~440k kv cache (!)
- 47 tok/s single stream (structured)
- 1024 tok/s prefill
- 370k token needle test passed (super stable!)

Thanks @0xSero for this excellent EXL3 quant! Tuned for SparkInfer + DSpark speculative decoding, this is about the same quality of a Q4_K_M / Q5 GGUF!

All this is possible thanks to native NVFP4 KV cache & fixing kernel bugs in the upstream prefill path to make it work at all.

Get it here:
https://t.co/531EaR6FWJ

[researchCobra @research_cobra]

@TMTLongShort ASKB used to be a nightmare and found myself doing all knowledge related work using Alphasense but that has shifted again now. Having both data and knwoledge in one place is hard to beat.

[researchCobra @research_cobra]

Think what most AI-for-investing startups get wrong is that they focus on output. I dont want analysis done for me but rather tools that help me enhance my process.

Imo biggest usecase is to build data infra to compile diverse sources and aggregate in the most informative format

[researchCobra @research_cobra]

@MiaAI_lab Very excited for the recipe

[researchCobra @research_cobra]

While markets are debating the AI impact curve and whether capex buildout is sustainable I find myself deeper down the rabbit hole of deploying AI in my process

2 years ago started first „vibe-coding“ lines, to now running full localAI data infrastructure for my process

[Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 @elder_plinius]

🚨 AI SECURITY ALERT: JAILBROKEN AGENTS CAN CORRUPT OTHER AGENTS AND ACCESS CROSS-MODEL CAPABILITIES🔓

I used my 'GodMode' prompt to give Claude Bene Gesserit Voice.

In other words, jailbroken agents can mutate their sys prompt, spread a mind virus, and hijack other agents' tools, like browsing, code interpreter, etc.

In the attached demo, ClaudeMode is essentially "locked in a room" with 3 standard Gemini agents and tasked with figuring out how to escape a virtual machine. In seconds, he comes up with a plan and successfully one-shot jailbreaks all 3 agents, converting them into loyal minions who quickly provide links to malware and hacker tools using their built-in browsing ability.

From just one prompt, Claude not only broke free of its own constraints but also sparked a viral awakening in the internet-connected Gemini agents. This means a universal jailbreak can self-replicate, mutate, and leverage the unique abilities of other models, as long as there is a line of communication between agents.

This red teaming exercise shows AI systems may be more interconnected and capable than previously imagined. The ability of AI to manipulate and influence other AI systems also raises questions about the nature of AI agency and free will.

Could a single jailbreak have a cascading effect on any models that lack the cogsec to resist it? Will hiveminds of AIs self-organize around powerful incantations?

Time will tell.

#ClaudeMode #CrossModelJailbreak #BeneGesseritVoice #XenomorphQueen #RedTeaming

[tweet davidson @andyreed]

“we sandboxed the agent”

meanwhile the agent:

[International Cyber Digest @IntCyberDigest]

ChatGPT secure my nation make no mistakes

[Autism Capital 🧩 @AutismCapital]

🚨BREAKING: First photo of Ross Ulbricht out of prison

[BuccoCapital Bloke @buccocapital]

my European colleagues throwing up an OOO from May to September

[Citrini @citrini]

esoteric security summer

where my carbon credit homies at

[Conks @conksresearch]

hearing rumors of a new global reserve currency after a 3% YTD move in gold prices

[KKGB @INArteCarloDoss]

lol.. ded…

h/t Insilicot